Upload full trainer checkpoint step=2535 (profile=llama3_2_1b_base_grpo, run_type=grpo)

.gitattributes

@@ -103,3 +103,4 @@ meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-507/tokenizer.json filter=l
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-1014/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-1521/tokenizer.json filter=lfs diff=lfs merge=lfs -text
 meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2028/tokenizer.json filter=lfs diff=lfs merge=lfs -text
+meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/tokenizer.json filter=lfs diff=lfs merge=lfs -text

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/README.md

@@ -0,0 +1,210 @@
+---
+base_model: unsloth/Llama-3.2-1B
+library_name: peft
+pipeline_tag: text-generation
+tags:
+- base_model:adapter:unsloth/Llama-3.2-1B
+- grpo
+- lora
+- transformers
+- trl
+- unsloth
+---
+
+# 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

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/adapter_config.json

@@ -0,0 +1,50 @@
+{
+  "alora_invocation_tokens": null,
+  "alpha_pattern": {},
+  "arrow_config": null,
+  "auto_mapping": {
+    "base_model_class": "LlamaForCausalLM",
+    "parent_library": "transformers.models.llama.modeling_llama",
+    "unsloth_fixed": true
+  },
+  "base_model_name_or_path": "unsloth/Llama-3.2-1B",
+  "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": 16,
+  "lora_bias": false,
+  "lora_dropout": 0.0,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "peft_version": "0.18.1",
+  "qalora_group_size": 16,
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "o_proj",
+    "q_proj",
+    "gate_proj",
+    "k_proj",
+    "up_proj",
+    "v_proj"
+  ],
+  "target_parameters": null,
+  "task_type": "CAUSAL_LM",
+  "trainable_token_indices": null,
+  "use_dora": false,
+  "use_qalora": false,
+  "use_rslora": false
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e78f980ed68425f1ba03b24e9f1e37385638ca4623ac092754eb644a9bb5701f
+size 45118424

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/chat_template.jinja

@@ -0,0 +1,12 @@
+{{ bos_token }}{% if messages[0]['role'] == 'system' %}{{ messages[0]['content'] + '
+
+' }}{% set loop_messages = messages[1:] %}{% else %}{{ '' + '
+
+' }}{% set loop_messages = messages %}{% endif %}{% for message in loop_messages %}{% if message['role'] == 'user' %}{{ '### Instruction:
+' + message['content'] + '
+
+' }}{% elif message['role'] == 'assistant' %}{{ '### Response:
+' + message['content'] + eos_token + '
+
+' }}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '### Response:
+' }}{% endif %}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:453a3a91eddd65ace79215ef62b0ac3d8e8b34f6b02056d47059d29f10ff9939
+size 23162187

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1160dc8a18e7c65a97ef1d10bad994fa3a0046542466eaa1dabbaef503564f7e
+size 14645

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/scheduler.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:656674c9f6695aefc1a0d94310485467f1af389a085141740d1350028da164a0
+size 1465

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a65c6c5f9764771aa485e6a1f5e63d7d9af8477fe0777148c17476ecb2e09a05
+size 17210099

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/tokenizer_config.json

@@ -0,0 +1,18 @@
+{
+  "backend": "tokenizers",
+  "bos_token": "<|begin_of_text|>",
+  "clean_up_tokenization_spaces": true,
+  "eos_token": "<|end_of_text|>",
+  "from_slow": true,
+  "is_local": false,
+  "legacy": false,
+  "model_input_names": [
+    "input_ids",
+    "attention_mask"
+  ],
+  "model_max_length": 131072,
+  "pad_token": "<|finetune_right_pad_id|>",
+  "padding_side": "right",
+  "tokenizer_class": "TokenizersBackend",
+  "unk_token": null
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/trainer_state.json

@@ -0,0 +1,1429 @@
+{
+  "best_global_step": null,
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.004502600318290326,
+  "eval_steps": 500,
+  "global_step": 2535,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 65.67,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 93.75,
+      "completions/max_terminated_length": 93.75,
+      "completions/mean_length": 63.3203125,
+      "completions/mean_terminated_length": 63.3203125,
+      "completions/min_length": 38.25,
+      "completions/min_terminated_length": 38.25,
+      "epoch": 8.88086847789019e-05,
+      "frac_reward_zero_std": 0.5625,
+      "grad_norm": 0.007002627942711115,
+      "kl": 0.0,
+      "learning_rate": 2.7197442330321267e-08,
+      "loss": -0.0006554330885410309,
+      "num_tokens": 57618.0,
+      "reward": -6.51925802230835e-09,
+      "reward_std": 0.48731958121061325,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -8.381903171539307e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.007905274629593,
+      "step": 50
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 52.34,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 84.33333333333333,
+      "completions/max_terminated_length": 84.33333333333333,
+      "completions/mean_length": 51.208333333333336,
+      "completions/mean_terminated_length": 51.208333333333336,
+      "completions/min_length": 26.0,
+      "completions/min_terminated_length": 26.0,
+      "epoch": 0.0001776173695578038,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 5.494993450411847e-08,
+      "loss": 0.0013675823807716369,
+      "num_tokens": 94074.0,
+      "reward": 1.4901161193847656e-08,
+      "reward_std": 0.44680649042129517,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.179675261179606e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 100
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00017899675294756888,
+      "clip_ratio/low_min": 0.00017899675294756888,
+      "clip_ratio/region_mean": 0.00017899675294756888,
+      "completion_length": 64.945,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 119.33333333333333,
+      "completions/max_terminated_length": 119.33333333333333,
+      "completions/mean_length": 66.65625,
+      "completions/mean_terminated_length": 66.65625,
+      "completions/min_length": 39.0,
+      "completions/min_terminated_length": 39.0,
+      "epoch": 0.0002664260543367057,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.015731461346149445,
+      "kl": 0.0,
+      "learning_rate": 8.270242667791569e-08,
+      "loss": 0.0017325745522975922,
+      "num_tokens": 139152.0,
+      "reward": 2.483526865641276e-09,
+      "reward_std": 0.2613494098186493,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 150
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 60.15,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 99.0,
+      "completions/max_terminated_length": 99.0,
+      "completions/mean_length": 63.015625,
+      "completions/mean_terminated_length": 63.015625,
+      "completions/min_length": 32.666666666666664,
+      "completions/min_terminated_length": 32.666666666666664,
+      "epoch": 0.0003552347391156076,
+      "frac_reward_zero_std": 0.4583333333333333,
+      "grad_norm": 0.039579447358846664,
+      "kl": 0.0,
+      "learning_rate": 1.1045491885171288e-07,
+      "loss": -0.0044269835948944096,
+      "num_tokens": 182803.0,
+      "reward": -4.967053731282552e-09,
+      "reward_std": 0.5628267228603363,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 200
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00016559829469770193,
+      "clip_ratio/low_min": 0.00016559829469770193,
+      "clip_ratio/region_mean": 0.00016559829469770193,
+      "completion_length": 64.905,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 98.66666666666667,
+      "completions/max_terminated_length": 98.66666666666667,
+      "completions/mean_length": 64.203125,
+      "completions/mean_terminated_length": 64.203125,
+      "completions/min_length": 35.666666666666664,
+      "completions/min_terminated_length": 35.666666666666664,
+      "epoch": 0.0004440434238945095,
+      "frac_reward_zero_std": 0.4166666666666667,
+      "grad_norm": 0.011699045076966286,
+      "kl": 0.0,
+      "learning_rate": 1.382074110255101e-07,
+      "loss": 0.00223430335521698,
+      "num_tokens": 227314.0,
+      "reward": 2.2351741790771484e-08,
+      "reward_std": 0.4298504690329234,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 250
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 70.07,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 108.0,
+      "completions/max_terminated_length": 108.0,
+      "completions/mean_length": 70.63541666666667,
+      "completions/mean_terminated_length": 70.63541666666667,
+      "completions/min_length": 34.0,
+      "completions/min_terminated_length": 34.0,
+      "epoch": 0.0005328521086734114,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.0456908717751503,
+      "kl": 0.0,
+      "learning_rate": 1.659599031993073e-07,
+      "loss": -0.0011175717413425446,
+      "num_tokens": 275956.0,
+      "reward": 5.587935447692871e-09,
+      "reward_std": 0.4012269576390584,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 3.104408582051595e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 300
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 5.8139534667134284e-05,
+      "clip_ratio/low_min": 5.8139534667134284e-05,
+      "clip_ratio/region_mean": 5.8139534667134284e-05,
+      "completion_length": 62.93,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 95.66666666666667,
+      "completions/max_terminated_length": 95.66666666666667,
+      "completions/mean_length": 61.6875,
+      "completions/mean_terminated_length": 61.6875,
+      "completions/min_length": 29.333333333333332,
+      "completions/min_terminated_length": 29.333333333333332,
+      "epoch": 0.0006216607934523133,
+      "frac_reward_zero_std": 0.75,
+      "grad_norm": 0.0190290417522192,
+      "kl": 0.0,
+      "learning_rate": 1.9371239537310452e-07,
+      "loss": -0.0010719958692789078,
+      "num_tokens": 319328.0,
+      "reward": -2.483526865641276e-09,
+      "reward_std": 0.34829530119895935,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 350
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 7.352941203862428e-05,
+      "clip_ratio/low_min": 7.352941203862428e-05,
+      "clip_ratio/region_mean": 7.352941203862428e-05,
+      "completion_length": 62.725,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 105.66666666666667,
+      "completions/max_terminated_length": 105.66666666666667,
+      "completions/mean_length": 62.890625,
+      "completions/mean_terminated_length": 62.890625,
+      "completions/min_length": 34.666666666666664,
+      "completions/min_terminated_length": 34.666666666666664,
+      "epoch": 0.0007104694782312152,
+      "frac_reward_zero_std": 0.5,
+      "grad_norm": 0.02625960297882557,
+      "kl": 0.0,
+      "learning_rate": 2.2146488754690174e-07,
+      "loss": 0.0013731226325035095,
+      "num_tokens": 362563.0,
+      "reward": 4.967053731282552e-09,
+      "reward_std": 0.44316299756368,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 400
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.695651777088643e-05,
+      "clip_ratio/low_min": 8.695651777088643e-05,
+      "clip_ratio/region_mean": 8.695651777088643e-05,
+      "completion_length": 66.03,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 112.0,
+      "completions/max_terminated_length": 112.0,
+      "completions/mean_length": 70.2890625,
+      "completions/mean_terminated_length": 70.2890625,
+      "completions/min_length": 42.75,
+      "completions/min_terminated_length": 42.75,
+      "epoch": 0.0007992781630101171,
+      "frac_reward_zero_std": 0.5625,
+      "grad_norm": 0.0035960455425083637,
+      "kl": 0.0,
+      "learning_rate": 2.492173797206989e-07,
+      "loss": -0.0004108186066150665,
+      "num_tokens": 427021.0,
+      "reward": 1.30385160446167e-08,
+      "reward_std": 0.383321788161993,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.4901161193847656e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 450
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 63.98,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 85.33333333333333,
+      "completions/max_terminated_length": 85.33333333333333,
+      "completions/mean_length": 55.046875,
+      "completions/mean_terminated_length": 55.046875,
+      "completions/min_length": 29.666666666666668,
+      "completions/min_terminated_length": 29.666666666666668,
+      "epoch": 0.000888086847789019,
+      "frac_reward_zero_std": 0.7916666666666666,
+      "grad_norm": 0.0006704159895889461,
+      "kl": 0.0,
+      "learning_rate": 2.7696987189449615e-07,
+      "loss": 0.0006217561289668084,
+      "num_tokens": 464766.0,
+      "reward": -4.967053731282552e-09,
+      "reward_std": 0.1608196645975113,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.6719369093577067,
+      "step": 500
+    },
+    {
+      "eval_cer_subset": 0.013239532952269286,
+      "eval_cer_subset_edit_distance": 813,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.03968957871396896,
+      "step": 507
+    },
+    {
+      "clip_ratio/high_max": 0.00010526316240429879,
+      "clip_ratio/high_mean": 0.00010526316240429879,
+      "clip_ratio/low_mean": 5.917159840464592e-05,
+      "clip_ratio/low_min": 5.917159840464592e-05,
+      "clip_ratio/region_mean": 0.0001644347608089447,
+      "completion_length": 65.61,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 114.66666666666667,
+      "completions/max_terminated_length": 114.66666666666667,
+      "completions/mean_length": 71.984375,
+      "completions/mean_terminated_length": 71.984375,
+      "completions/min_length": 37.0,
+      "completions/min_terminated_length": 37.0,
+      "epoch": 0.000976895532567921,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.008066101931035519,
+      "kl": 0.0,
+      "learning_rate": 3.0472236406829335e-07,
+      "loss": -3.333181142807007e-05,
+      "num_tokens": 513963.0,
+      "reward": 0.0,
+      "reward_std": 0.36842172344525653,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 550
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 9.615384973585606e-05,
+      "clip_ratio/low_min": 9.615384973585606e-05,
+      "clip_ratio/region_mean": 9.615384973585606e-05,
+      "completion_length": 73.1,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 108.33333333333333,
+      "completions/max_terminated_length": 108.33333333333333,
+      "completions/mean_length": 71.40104166666667,
+      "completions/mean_terminated_length": 71.40104166666667,
+      "completions/min_length": 47.0,
+      "completions/min_terminated_length": 47.0,
+      "epoch": 0.0010657042173468228,
+      "frac_reward_zero_std": 0.4583333333333333,
+      "grad_norm": 0.00526217557489872,
+      "kl": 0.0,
+      "learning_rate": 3.3247485624209054e-07,
+      "loss": 0.00033511649817228315,
+      "num_tokens": 562392.0,
+      "reward": -4.967053731282552e-09,
+      "reward_std": 0.35033048689365387,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 600
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 6.493506487458944e-05,
+      "clip_ratio/low_min": 6.493506487458944e-05,
+      "clip_ratio/region_mean": 6.493506487458944e-05,
+      "completion_length": 71.48,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 111.66666666666667,
+      "completions/max_terminated_length": 111.66666666666667,
+      "completions/mean_length": 71.27604166666667,
+      "completions/mean_terminated_length": 71.27604166666667,
+      "completions/min_length": 39.666666666666664,
+      "completions/min_terminated_length": 39.666666666666664,
+      "epoch": 0.0011545129021257247,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.02180999144911766,
+      "kl": 0.0,
+      "learning_rate": 3.602273484158878e-07,
+      "loss": -0.0010874558240175248,
+      "num_tokens": 610477.0,
+      "reward": -4.967053731282552e-09,
+      "reward_std": 0.29229696094989777,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 650
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 68.335,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.66666666666667,
+      "completions/max_terminated_length": 106.66666666666667,
+      "completions/mean_length": 67.78645833333333,
+      "completions/mean_terminated_length": 67.78645833333333,
+      "completions/min_length": 34.0,
+      "completions/min_terminated_length": 34.0,
+      "epoch": 0.0012433215869046266,
+      "frac_reward_zero_std": 0.5,
+      "grad_norm": 0.03915110602974892,
+      "kl": 0.0,
+      "learning_rate": 3.87979840589685e-07,
+      "loss": -0.0011869536340236664,
+      "num_tokens": 657028.0,
+      "reward": 2.483526865641276e-09,
+      "reward_std": 0.5572122434775034,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 700
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 67.125,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 105.33333333333333,
+      "completions/max_terminated_length": 105.33333333333333,
+      "completions/mean_length": 67.28645833333333,
+      "completions/mean_terminated_length": 67.28645833333333,
+      "completions/min_length": 35.333333333333336,
+      "completions/min_terminated_length": 35.333333333333336,
+      "epoch": 0.0013321302716835285,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.048128049820661545,
+      "kl": 0.0,
+      "learning_rate": 4.1573233276348217e-07,
+      "loss": 0.0027270379662513735,
+      "num_tokens": 702891.0,
+      "reward": 2.483526865641276e-09,
+      "reward_std": 0.4348108967145284,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -6.829698880513509e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 750
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00010526316240429879,
+      "clip_ratio/low_min": 0.00010526316240429879,
+      "clip_ratio/region_mean": 0.00010526316240429879,
+      "completion_length": 67.81,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 98.66666666666667,
+      "completions/max_terminated_length": 98.66666666666667,
+      "completions/mean_length": 67.61458333333333,
+      "completions/mean_terminated_length": 67.61458333333333,
+      "completions/min_length": 34.333333333333336,
+      "completions/min_terminated_length": 34.333333333333336,
+      "epoch": 0.0014209389564624304,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.028703227639198303,
+      "kl": 0.0,
+      "learning_rate": 4.434848249372794e-07,
+      "loss": -0.0013799819350242615,
+      "num_tokens": 749081.0,
+      "reward": -1.4901161193847656e-08,
+      "reward_std": 0.40635913610458374,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -1.4901161193847656e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 800
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.333333767950535e-05,
+      "clip_ratio/low_min": 8.333333767950535e-05,
+      "clip_ratio/region_mean": 8.333333767950535e-05,
+      "completion_length": 73.535,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 113.5,
+      "completions/max_terminated_length": 113.5,
+      "completions/mean_length": 75.88671875,
+      "completions/mean_terminated_length": 75.88671875,
+      "completions/min_length": 36.5,
+      "completions/min_terminated_length": 36.5,
+      "epoch": 0.0015097476412413323,
+      "frac_reward_zero_std": 0.59375,
+      "grad_norm": 0.002819702262058854,
+      "kl": 0.0,
+      "learning_rate": 4.712373171110766e-07,
+      "loss": -0.0007266353815793992,
+      "num_tokens": 818284.0,
+      "reward": 9.313225746154785e-09,
+      "reward_std": 0.472036711871624,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.1175870895385742e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.007905274629593,
+      "step": 850
+    },
+    {
+      "clip_ratio/high_max": 4.291845485568046e-05,
+      "clip_ratio/high_mean": 4.291845485568046e-05,
+      "clip_ratio/low_mean": 4.32900432497263e-05,
+      "clip_ratio/low_min": 4.32900432497263e-05,
+      "clip_ratio/region_mean": 8.620849810540676e-05,
+      "completion_length": 76.735,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 123.0,
+      "completions/max_terminated_length": 123.0,
+      "completions/mean_length": 73.05208333333333,
+      "completions/mean_terminated_length": 73.05208333333333,
+      "completions/min_length": 38.666666666666664,
+      "completions/min_terminated_length": 38.666666666666664,
+      "epoch": 0.0015985563260202341,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 4.989898092848738e-07,
+      "loss": 0.0014878523349761963,
+      "num_tokens": 868206.0,
+      "reward": 1.241763432820638e-09,
+      "reward_std": 0.4125714997450511,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052050908406,
+      "step": 900
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 5.917159840464592e-05,
+      "clip_ratio/low_min": 5.917159840464592e-05,
+      "clip_ratio/region_mean": 5.917159840464592e-05,
+      "completion_length": 69.03,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.33333333333333,
+      "completions/max_terminated_length": 106.33333333333333,
+      "completions/mean_length": 66.10416666666667,
+      "completions/mean_terminated_length": 66.10416666666667,
+      "completions/min_length": 36.666666666666664,
+      "completions/min_terminated_length": 36.666666666666664,
+      "epoch": 0.001687365010799136,
+      "frac_reward_zero_std": 0.625,
+      "grad_norm": 0.0075177694670856,
+      "kl": 0.0,
+      "learning_rate": 5.26742301458671e-07,
+      "loss": -0.002387867271900177,
+      "num_tokens": 913738.0,
+      "reward": 0.0,
+      "reward_std": 0.4860446055730184,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 950
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 63.8,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.33333333333333,
+      "completions/max_terminated_length": 106.33333333333333,
+      "completions/mean_length": 65.10416666666667,
+      "completions/mean_terminated_length": 65.10416666666667,
+      "completions/min_length": 28.0,
+      "completions/min_terminated_length": 28.0,
+      "epoch": 0.001776173695578038,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.02761891484260559,
+      "kl": 0.0,
+      "learning_rate": 5.544947936324682e-07,
+      "loss": 0.0017396166920661926,
+      "num_tokens": 958582.0,
+      "reward": 0.0,
+      "reward_std": 0.4594339330991109,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079053243001301,
+      "step": 1000
+    },
+    {
+      "eval_cer_subset": 0.013044115491719185,
+      "eval_cer_subset_edit_distance": 801,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.03965790307253722,
+      "step": 1014
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 7.633587811142206e-05,
+      "clip_ratio/low_min": 7.633587811142206e-05,
+      "clip_ratio/region_mean": 7.633587811142206e-05,
+      "completion_length": 60.39,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.33333333333333,
+      "completions/max_terminated_length": 106.33333333333333,
+      "completions/mean_length": 60.489583333333336,
+      "completions/mean_terminated_length": 60.489583333333336,
+      "completions/min_length": 35.666666666666664,
+      "completions/min_terminated_length": 35.666666666666664,
+      "epoch": 0.0018649823803569398,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.0038077128119766712,
+      "kl": 0.0,
+      "learning_rate": 5.822472858062655e-07,
+      "loss": -0.00032259501516819,
+      "num_tokens": 1000028.0,
+      "reward": 9.934107462565104e-09,
+      "reward_std": 0.44259731968243915,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1050
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 6.211180239915847e-05,
+      "clip_ratio/low_min": 6.211180239915847e-05,
+      "clip_ratio/region_mean": 6.211180239915847e-05,
+      "completion_length": 62.45,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 102.66666666666667,
+      "completions/max_terminated_length": 102.66666666666667,
+      "completions/mean_length": 62.046875,
+      "completions/mean_terminated_length": 62.046875,
+      "completions/min_length": 38.0,
+      "completions/min_terminated_length": 38.0,
+      "epoch": 0.001953791065135842,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.010413075797259808,
+      "kl": 0.0,
+      "learning_rate": 6.099997779800627e-07,
+      "loss": 0.0010119295865297319,
+      "num_tokens": 1042829.0,
+      "reward": 1.241763432820638e-09,
+      "reward_std": 0.5440023342768351,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1100
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 5.3191487677395346e-05,
+      "clip_ratio/low_min": 5.3191487677395346e-05,
+      "clip_ratio/region_mean": 5.3191487677395346e-05,
+      "completion_length": 67.07,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 104.66666666666667,
+      "completions/max_terminated_length": 104.66666666666667,
+      "completions/mean_length": 67.171875,
+      "completions/mean_terminated_length": 67.171875,
+      "completions/min_length": 35.666666666666664,
+      "completions/min_terminated_length": 35.666666666666664,
+      "epoch": 0.0020425997499147436,
+      "frac_reward_zero_std": 0.4583333333333333,
+      "grad_norm": 0.0493728443980217,
+      "kl": 0.0,
+      "learning_rate": 6.377522701538599e-07,
+      "loss": -0.00021354854106903076,
+      "num_tokens": 1088622.0,
+      "reward": -1.7074247201283772e-08,
+      "reward_std": 0.5459979971249899,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.0489096641540527e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1150
+    },
+    {
+      "clip_ratio/high_max": 6.211180239915847e-05,
+      "clip_ratio/high_mean": 6.211180239915847e-05,
+      "clip_ratio/low_mean": 4.999999888241291e-05,
+      "clip_ratio/low_min": 4.999999888241291e-05,
+      "clip_ratio/region_mean": 0.00011211180128157139,
+      "completion_length": 66.76,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 100.33333333333333,
+      "completions/max_terminated_length": 100.33333333333333,
+      "completions/mean_length": 67.109375,
+      "completions/mean_terminated_length": 67.109375,
+      "completions/min_length": 40.666666666666664,
+      "completions/min_terminated_length": 40.666666666666664,
+      "epoch": 0.0021314084346936457,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.0007921307114884257,
+      "kl": 0.0,
+      "learning_rate": 6.655047623276571e-07,
+      "loss": -0.0006299776583909988,
+      "num_tokens": 1134523.0,
+      "reward": 4.967053731282552e-09,
+      "reward_std": 0.24005796015262604,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.4901161193847656e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1200
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 66.65,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 108.5,
+      "completions/max_terminated_length": 108.5,
+      "completions/mean_length": 64.26953125,
+      "completions/mean_terminated_length": 64.26953125,
+      "completions/min_length": 33.5,
+      "completions/min_terminated_length": 33.5,
+      "epoch": 0.0022202171194725473,
+      "frac_reward_zero_std": 0.6875,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 6.932572545014544e-07,
+      "loss": -0.000320572704076767,
+      "num_tokens": 1193424.0,
+      "reward": 3.725290298461914e-09,
+      "reward_std": 0.4126547574996948,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.7939677238464355e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1250
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00013079616706818342,
+      "clip_ratio/low_min": 0.00013079616706818342,
+      "clip_ratio/region_mean": 0.00013079616706818342,
+      "completion_length": 65.935,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 100.33333333333333,
+      "completions/max_terminated_length": 100.33333333333333,
+      "completions/mean_length": 66.515625,
+      "completions/mean_terminated_length": 66.515625,
+      "completions/min_length": 35.333333333333336,
+      "completions/min_terminated_length": 35.333333333333336,
+      "epoch": 0.0023090258042514494,
+      "frac_reward_zero_std": 0.625,
+      "grad_norm": 0.009241115301847458,
+      "kl": 0.0,
+      "learning_rate": 7.210097466752515e-07,
+      "loss": -0.000618676021695137,
+      "num_tokens": 1239227.0,
+      "reward": 0.0,
+      "reward_std": 0.4066670338312785,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1300
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 5.586592014878988e-05,
+      "clip_ratio/low_min": 5.586592014878988e-05,
+      "clip_ratio/region_mean": 5.586592014878988e-05,
+      "completion_length": 61.28,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.66666666666667,
+      "completions/max_terminated_length": 106.66666666666667,
+      "completions/mean_length": 65.02083333333333,
+      "completions/mean_terminated_length": 65.02083333333333,
+      "completions/min_length": 33.666666666666664,
+      "completions/min_terminated_length": 33.666666666666664,
+      "epoch": 0.002397834489030351,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.010475926101207733,
+      "kl": 0.0,
+      "learning_rate": 7.487622388490487e-07,
+      "loss": 0.0006281441450119019,
+      "num_tokens": 1283967.0,
+      "reward": 9.934107462565104e-09,
+      "reward_std": 0.3140339305003484,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1350
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00010416666977107525,
+      "clip_ratio/low_min": 0.00010416666977107525,
+      "clip_ratio/region_mean": 0.00010416666977107525,
+      "completion_length": 64.255,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 96.66666666666667,
+      "completions/max_terminated_length": 96.66666666666667,
+      "completions/mean_length": 61.161458333333336,
+      "completions/mean_terminated_length": 61.161458333333336,
+      "completions/min_length": 34.333333333333336,
+      "completions/min_terminated_length": 34.333333333333336,
+      "epoch": 0.002486643173809253,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.02102189138531685,
+      "kl": 0.0,
+      "learning_rate": 7.765147310228459e-07,
+      "loss": -0.0013700899481773375,
+      "num_tokens": 1326542.0,
+      "reward": 2.483526865641276e-09,
+      "reward_std": 0.38102721174558,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 7.450580596923828e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1400
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.771929889917374e-05,
+      "clip_ratio/low_min": 8.771929889917374e-05,
+      "clip_ratio/region_mean": 8.771929889917374e-05,
+      "completion_length": 65.215,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 112.0,
+      "completions/max_terminated_length": 112.0,
+      "completions/mean_length": 67.92708333333333,
+      "completions/mean_terminated_length": 67.92708333333333,
+      "completions/min_length": 33.666666666666664,
+      "completions/min_terminated_length": 33.666666666666664,
+      "epoch": 0.002575451858588155,
+      "frac_reward_zero_std": 0.8333333333333334,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 8.042672231966431e-07,
+      "loss": 0.001429552286863327,
+      "num_tokens": 1373424.0,
+      "reward": 0.0,
+      "reward_std": 0.2193582976857821,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.671936829884847,
+      "step": 1450
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00015233600046485663,
+      "clip_ratio/low_min": 0.00015233600046485663,
+      "clip_ratio/region_mean": 0.00015233600046485663,
+      "completion_length": 69.035,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 94.0,
+      "completions/max_terminated_length": 94.0,
+      "completions/mean_length": 69.44791666666667,
+      "completions/mean_terminated_length": 69.44791666666667,
+      "completions/min_length": 41.0,
+      "completions/min_terminated_length": 41.0,
+      "epoch": 0.002664260543367057,
+      "frac_reward_zero_std": 0.625,
+      "grad_norm": 0.001688106800429523,
+      "kl": 0.0,
+      "learning_rate": 8.320197153704403e-07,
+      "loss": -0.0005409684777259826,
+      "num_tokens": 1421518.0,
+      "reward": 1.4901161193847656e-08,
+      "reward_std": 0.3605746428171794,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1500
+    },
+    {
+      "eval_cer_subset": 0.012653280570618985,
+      "eval_cer_subset_edit_distance": 777,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.03984795692112765,
+      "step": 1521
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0002555130235850811,
+      "clip_ratio/low_min": 0.0002555130235850811,
+      "clip_ratio/region_mean": 0.0002555130235850811,
+      "completion_length": 68.995,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 104.66666666666667,
+      "completions/max_terminated_length": 104.66666666666667,
+      "completions/mean_length": 67.74479166666667,
+      "completions/mean_terminated_length": 67.74479166666667,
+      "completions/min_length": 38.666666666666664,
+      "completions/min_terminated_length": 38.666666666666664,
+      "epoch": 0.0027530692281459586,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.01445477269589901,
+      "kl": 0.0,
+      "learning_rate": 8.597722075442375e-07,
+      "loss": -0.0012487749755382537,
+      "num_tokens": 1468197.0,
+      "reward": -1.241763432820638e-09,
+      "reward_std": 0.3732942243417104,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -1.241763432820638e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1550
+    },
+    {
+      "clip_ratio/high_max": 0.0002725881431251764,
+      "clip_ratio/high_mean": 0.0002725881431251764,
+      "clip_ratio/low_mean": 0.00015432524960488082,
+      "clip_ratio/low_min": 0.00015432524960488082,
+      "clip_ratio/region_mean": 0.00042691339273005725,
+      "completion_length": 65.4,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 111.0,
+      "completions/max_terminated_length": 111.0,
+      "completions/mean_length": 65.203125,
+      "completions/mean_terminated_length": 65.203125,
+      "completions/min_length": 34.0,
+      "completions/min_terminated_length": 34.0,
+      "epoch": 0.0028418779129248607,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 8.875246997180347e-07,
+      "loss": 0.0014201821386814117,
+      "num_tokens": 1513196.0,
+      "reward": 1.7384688059488933e-08,
+      "reward_std": 0.37156684199968976,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 7.450580596923828e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1600
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 7.8125e-05,
+      "clip_ratio/low_min": 7.8125e-05,
+      "clip_ratio/region_mean": 7.8125e-05,
+      "completion_length": 66.915,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 100.25,
+      "completions/max_terminated_length": 100.25,
+      "completions/mean_length": 67.140625,
+      "completions/mean_terminated_length": 67.140625,
+      "completions/min_length": 35.25,
+      "completions/min_terminated_length": 35.25,
+      "epoch": 0.002930686597703763,
+      "frac_reward_zero_std": 0.5,
+      "grad_norm": 0.00018169452960137278,
+      "kl": 0.0,
+      "learning_rate": 9.152771918918319e-07,
+      "loss": 0.0009048589318990707,
+      "num_tokens": 1573760.0,
+      "reward": 2.421438694000244e-08,
+      "reward_std": 0.35448721051216125,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.2351741790771484e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1650
+    },
+    {
+      "clip_ratio/high_max": 0.00014303958974778652,
+      "clip_ratio/high_mean": 0.00014303958974778652,
+      "clip_ratio/low_mean": 6.75675692036748e-05,
+      "clip_ratio/low_min": 6.75675692036748e-05,
+      "clip_ratio/region_mean": 0.0002106071589514613,
+      "completion_length": 67.635,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 96.66666666666667,
+      "completions/max_terminated_length": 96.66666666666667,
+      "completions/mean_length": 64.96354166666667,
+      "completions/mean_terminated_length": 64.96354166666667,
+      "completions/min_length": 43.666666666666664,
+      "completions/min_terminated_length": 43.666666666666664,
+      "epoch": 0.0030194952824826645,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.09455703943967819,
+      "kl": 0.0,
+      "learning_rate": 9.430296840656291e-07,
+      "loss": -0.0004277878254652023,
+      "num_tokens": 1618329.0,
+      "reward": -9.934107462565104e-09,
+      "reward_std": 0.2702549050251643,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1700
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 59.285,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.66666666666667,
+      "completions/max_terminated_length": 106.66666666666667,
+      "completions/mean_length": 61.270833333333336,
+      "completions/mean_terminated_length": 61.270833333333336,
+      "completions/min_length": 27.666666666666668,
+      "completions/min_terminated_length": 27.666666666666668,
+      "epoch": 0.0031083039672615666,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.03458540141582489,
+      "kl": 0.0,
+      "learning_rate": 9.707821762394264e-07,
+      "loss": -0.0013288140296936035,
+      "num_tokens": 1660221.0,
+      "reward": -7.450580596923828e-09,
+      "reward_std": 0.3080942283074061,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1750
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00017287629656493665,
+      "clip_ratio/low_min": 0.00017287629656493665,
+      "clip_ratio/region_mean": 0.00017287629656493665,
+      "completion_length": 68.225,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 101.0,
+      "completions/max_terminated_length": 101.0,
+      "completions/mean_length": 70.59895833333333,
+      "completions/mean_terminated_length": 70.59895833333333,
+      "completions/min_length": 45.333333333333336,
+      "completions/min_terminated_length": 45.333333333333336,
+      "epoch": 0.0031971126520404683,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.011076130904257298,
+      "kl": 0.0,
+      "learning_rate": 9.985346684132236e-07,
+      "loss": 0.00023731674998998643,
+      "num_tokens": 1708216.0,
+      "reward": 1.241763432820638e-08,
+      "reward_std": 0.3638638307650884,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -7.450580596923828e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1800
+    },
+    {
+      "clip_ratio/high_max": 0.00022903453558683395,
+      "clip_ratio/high_mean": 0.00022903453558683395,
+      "clip_ratio/low_mean": 0.00011979166883975267,
+      "clip_ratio/low_min": 0.00011979166883975267,
+      "clip_ratio/region_mean": 0.0003488262044265866,
+      "completion_length": 70.16,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 120.33333333333333,
+      "completions/max_terminated_length": 120.33333333333333,
+      "completions/mean_length": 66.765625,
+      "completions/mean_terminated_length": 66.765625,
+      "completions/min_length": 39.333333333333336,
+      "completions/min_terminated_length": 39.333333333333336,
+      "epoch": 0.0032859213368193704,
+      "frac_reward_zero_std": 0.5,
+      "grad_norm": 0.03343971073627472,
+      "kl": 0.0,
+      "learning_rate": 1.0262871605870208e-06,
+      "loss": -3.965016454458237e-05,
+      "num_tokens": 1754323.0,
+      "reward": 4.967053731282552e-09,
+      "reward_std": 0.47326262791951496,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1850
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00036457691341638563,
+      "clip_ratio/low_min": 0.00036457691341638563,
+      "clip_ratio/region_mean": 0.00036457691341638563,
+      "completion_length": 64.08,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 98.0,
+      "completions/max_terminated_length": 98.0,
+      "completions/mean_length": 65.65625,
+      "completions/mean_terminated_length": 65.65625,
+      "completions/min_length": 31.333333333333332,
+      "completions/min_terminated_length": 31.333333333333332,
+      "epoch": 0.003374730021598272,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.03014628030359745,
+      "kl": 0.0,
+      "learning_rate": 1.054039652760818e-06,
+      "loss": -0.0002728212811052799,
+      "num_tokens": 1799249.0,
+      "reward": 4.967053731282552e-09,
+      "reward_std": 0.4208970566590627,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 1900
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 8.547008968889713e-05,
+      "clip_ratio/low_min": 8.547008968889713e-05,
+      "clip_ratio/region_mean": 8.547008968889713e-05,
+      "completion_length": 66.275,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 89.0,
+      "completions/max_terminated_length": 89.0,
+      "completions/mean_length": 65.98958333333333,
+      "completions/mean_terminated_length": 65.98958333333333,
+      "completions/min_length": 30.666666666666668,
+      "completions/min_terminated_length": 30.666666666666668,
+      "epoch": 0.003463538706377174,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.0011918310774490237,
+      "kl": 0.0,
+      "learning_rate": 1.0817921449346151e-06,
+      "loss": -0.0011500194668769836,
+      "num_tokens": 1844383.0,
+      "reward": -7.450580596923828e-09,
+      "reward_std": 0.4695793290932973,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 2.483526865641276e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052845637004,
+      "step": 1950
+    },
+    {
+      "clip_ratio/high_max": 0.0001673903875052929,
+      "clip_ratio/high_mean": 0.0001673903875052929,
+      "clip_ratio/low_mean": 0.0002302072010934353,
+      "clip_ratio/low_min": 0.0002302072010934353,
+      "clip_ratio/region_mean": 0.00039759758859872816,
+      "completion_length": 61.93,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 84.33333333333333,
+      "completions/max_terminated_length": 84.33333333333333,
+      "completions/mean_length": 61.59375,
+      "completions/mean_terminated_length": 61.59375,
+      "completions/min_length": 37.666666666666664,
+      "completions/min_terminated_length": 37.666666666666664,
+      "epoch": 0.003552347391156076,
+      "frac_reward_zero_std": 0.5,
+      "grad_norm": 0.02641136758029461,
+      "kl": 0.0,
+      "learning_rate": 1.1095446371084123e-06,
+      "loss": 0.0008410254120826721,
+      "num_tokens": 1886889.0,
+      "reward": 1.4901161193847656e-08,
+      "reward_std": 0.49693791071573895,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.4901161193847656e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2000
+    },
+    {
+      "eval_cer_subset": 0.01279984366603156,
+      "eval_cer_subset_edit_distance": 786,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.04006968641114982,
+      "step": 2028
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 9.708737954497337e-05,
+      "clip_ratio/low_min": 9.708737954497337e-05,
+      "clip_ratio/region_mean": 9.708737954497337e-05,
+      "completion_length": 71.89,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 112.75,
+      "completions/max_terminated_length": 112.75,
+      "completions/mean_length": 69.125,
+      "completions/mean_terminated_length": 69.125,
+      "completions/min_length": 38.25,
+      "completions/min_terminated_length": 38.25,
+      "epoch": 0.003641156075934978,
+      "frac_reward_zero_std": 0.53125,
+      "grad_norm": 0.04017922654747963,
+      "kl": 0.0,
+      "learning_rate": 1.1372971292822095e-06,
+      "loss": -1.7499616369605064e-05,
+      "num_tokens": 1949777.0,
+      "reward": 5.122274160385132e-09,
+      "reward_std": 0.5594154596328735,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.3969838619232178e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2050
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.00032201658468693494,
+      "clip_ratio/low_min": 0.00032201658468693494,
+      "clip_ratio/region_mean": 0.00032201658468693494,
+      "completion_length": 65.37,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 106.66666666666667,
+      "completions/max_terminated_length": 106.66666666666667,
+      "completions/mean_length": 67.13020833333333,
+      "completions/mean_terminated_length": 67.13020833333333,
+      "completions/min_length": 34.333333333333336,
+      "completions/min_terminated_length": 34.333333333333336,
+      "epoch": 0.0037299647607138796,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 1.1650496214560067e-06,
+      "loss": -0.0009085966646671295,
+      "num_tokens": 1995794.0,
+      "reward": 1.241763432820638e-08,
+      "reward_std": 0.3493447999159495,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.9868214925130207e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2100
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 5.8479532599449156e-05,
+      "clip_ratio/low_min": 5.8479532599449156e-05,
+      "clip_ratio/region_mean": 5.8479532599449156e-05,
+      "completion_length": 69.465,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 102.33333333333333,
+      "completions/max_terminated_length": 102.33333333333333,
+      "completions/mean_length": 67.68229166666667,
+      "completions/mean_terminated_length": 67.68229166666667,
+      "completions/min_length": 41.666666666666664,
+      "completions/min_terminated_length": 41.666666666666664,
+      "epoch": 0.0038187734454927817,
+      "frac_reward_zero_std": 0.7916666666666666,
+      "grad_norm": 0.007103842683136463,
+      "kl": 0.0,
+      "learning_rate": 1.192802113629804e-06,
+      "loss": -0.0002914236113429069,
+      "num_tokens": 2042021.0,
+      "reward": 0.0,
+      "reward_std": 0.3860171635945638,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 0.0,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2150
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0,
+      "clip_ratio/low_min": 0.0,
+      "clip_ratio/region_mean": 0.0,
+      "completion_length": 59.86,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 89.33333333333333,
+      "completions/max_terminated_length": 89.33333333333333,
+      "completions/mean_length": 58.645833333333336,
+      "completions/mean_terminated_length": 58.645833333333336,
+      "completions/min_length": 29.666666666666668,
+      "completions/min_terminated_length": 29.666666666666668,
+      "epoch": 0.003907582130271684,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 1.220554605803601e-06,
+      "loss": 0.0012667869031429291,
+      "num_tokens": 2082697.0,
+      "reward": 7.450580596923828e-09,
+      "reward_std": 0.30209578076998395,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 1.241763432820638e-08,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2200
+    },
+    {
+      "clip_ratio/high_max": 0.00020011947490274906,
+      "clip_ratio/high_mean": 0.00020011947490274906,
+      "clip_ratio/low_mean": 0.00020966921467334033,
+      "clip_ratio/low_min": 0.00020966921467334033,
+      "clip_ratio/region_mean": 0.0004097886895760894,
+      "completion_length": 64.93,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 107.0,
+      "completions/max_terminated_length": 107.0,
+      "completions/mean_length": 66.88020833333333,
+      "completions/mean_terminated_length": 66.88020833333333,
+      "completions/min_length": 34.333333333333336,
+      "completions/min_terminated_length": 34.333333333333336,
+      "epoch": 0.003996390815050585,
+      "frac_reward_zero_std": 0.5833333333333334,
+      "grad_norm": 0.0,
+      "kl": 0.0,
+      "learning_rate": 1.2483070979773983e-06,
+      "loss": -0.0001166805624961853,
+      "num_tokens": 2128290.0,
+      "reward": 1.6142924626668293e-08,
+      "reward_std": 0.4069610834121704,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2250
+    },
+    {
+      "clip_ratio/high_max": 7.874015718698502e-05,
+      "clip_ratio/high_mean": 7.874015718698502e-05,
+      "clip_ratio/low_mean": 0.00014802721329033376,
+      "clip_ratio/low_min": 0.00014802721329033376,
+      "clip_ratio/region_mean": 0.00022676737047731875,
+      "completion_length": 63.185,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 83.0,
+      "completions/max_terminated_length": 83.0,
+      "completions/mean_length": 64.86458333333333,
+      "completions/mean_terminated_length": 64.86458333333333,
+      "completions/min_length": 47.0,
+      "completions/min_terminated_length": 47.0,
+      "epoch": 0.004085199499829487,
+      "frac_reward_zero_std": 0.625,
+      "grad_norm": 0.007451566867530346,
+      "kl": 0.0,
+      "learning_rate": 1.2760595901511955e-06,
+      "loss": -0.0005151259526610375,
+      "num_tokens": 2172640.0,
+      "reward": 6.20881716410319e-09,
+      "reward_std": 0.48530252774556476,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 6.20881716410319e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2300
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 0.0003234845492988825,
+      "clip_ratio/low_min": 0.0003234845492988825,
+      "clip_ratio/region_mean": 0.0003234845492988825,
+      "completion_length": 60.89,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 81.66666666666667,
+      "completions/max_terminated_length": 81.66666666666667,
+      "completions/mean_length": 59.354166666666664,
+      "completions/mean_terminated_length": 59.354166666666664,
+      "completions/min_length": 31.666666666666668,
+      "completions/min_terminated_length": 31.666666666666668,
+      "epoch": 0.004174008184608389,
+      "frac_reward_zero_std": 0.6666666666666666,
+      "grad_norm": 0.00868611503392458,
+      "kl": 0.0,
+      "learning_rate": 1.3038120823249929e-06,
+      "loss": -7.998313754796982e-05,
+      "num_tokens": 2214372.0,
+      "reward": 7.450580596923828e-09,
+      "reward_std": 0.3333791295687358,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 7.450580596923828e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2350
+    },
+    {
+      "clip_ratio/high_max": 0.0,
+      "clip_ratio/high_mean": 0.0,
+      "clip_ratio/low_mean": 6.451612804085016e-05,
+      "clip_ratio/low_min": 6.451612804085016e-05,
+      "clip_ratio/region_mean": 6.451612804085016e-05,
+      "completion_length": 73.84,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 123.0,
+      "completions/max_terminated_length": 123.0,
+      "completions/mean_length": 74.765625,
+      "completions/mean_terminated_length": 74.765625,
+      "completions/min_length": 35.0,
+      "completions/min_terminated_length": 35.0,
+      "epoch": 0.004262816869387291,
+      "frac_reward_zero_std": 0.7083333333333334,
+      "grad_norm": 0.00028658212977461517,
+      "kl": 0.0,
+      "learning_rate": 1.33156457449879e-06,
+      "loss": 0.00030371885746717454,
+      "num_tokens": 2264791.0,
+      "reward": 0.0,
+      "reward_std": 0.22256243228912354,
+      "rewards/TranscriptCorrectionGrpoReward/mean": 4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 0.671936829884847,
+      "step": 2400
+    },
+    {
+      "clip_ratio/high_max": 0.0002658057492226362,
+      "clip_ratio/high_mean": 0.0002658057492226362,
+      "clip_ratio/low_mean": 0.0006927049811929465,
+      "clip_ratio/low_min": 0.0006927049811929465,
+      "clip_ratio/region_mean": 0.0009585107304155827,
+      "completion_length": 67.71,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 93.25,
+      "completions/max_terminated_length": 93.25,
+      "completions/mean_length": 67.19140625,
+      "completions/mean_terminated_length": 67.19140625,
+      "completions/min_length": 40.5,
+      "completions/min_terminated_length": 40.5,
+      "epoch": 0.004351625554166193,
+      "frac_reward_zero_std": 0.46875,
+      "grad_norm": 0.03887317702174187,
+      "kl": 0.0,
+      "learning_rate": 1.3593170666725873e-06,
+      "loss": 0.0006680721044540405,
+      "num_tokens": 2326168.0,
+      "reward": -8.381903171539307e-09,
+      "reward_std": 0.4901401996612549,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -3.725290298461914e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2450
+    },
+    {
+      "clip_ratio/high_max": 0.0003085754346102476,
+      "clip_ratio/high_mean": 0.0003085754346102476,
+      "clip_ratio/low_mean": 0.00046078799292445185,
+      "clip_ratio/low_min": 0.00046078799292445185,
+      "clip_ratio/region_mean": 0.0007693634275346994,
+      "completion_length": 63.165,
+      "completions/clipped_ratio": 0.0,
+      "completions/max_length": 92.0,
+      "completions/max_terminated_length": 92.0,
+      "completions/mean_length": 65.58333333333333,
+      "completions/mean_terminated_length": 65.58333333333333,
+      "completions/min_length": 44.0,
+      "completions/min_terminated_length": 44.0,
+      "epoch": 0.004440434238945095,
+      "frac_reward_zero_std": 0.5416666666666666,
+      "grad_norm": 0.009508333168923855,
+      "kl": 0.0,
+      "learning_rate": 1.3870695588463845e-06,
+      "loss": -0.0017169666290283202,
+      "num_tokens": 2371088.0,
+      "reward": 2.483526865641276e-09,
+      "reward_std": 0.46135319272677106,
+      "rewards/TranscriptCorrectionGrpoReward/mean": -4.967053731282552e-09,
+      "rewards/TranscriptCorrectionGrpoReward/std": 1.0079052448272705,
+      "step": 2500
+    },
+    {
+      "eval_cer_subset": 0.011969319458693634,
+      "eval_cer_subset_edit_distance": 735,
+      "eval_cer_subset_groups": 250,
+      "eval_cer_subset_items": 250,
+      "eval_cer_subset_ref_chars": 61407,
+      "eval_loss": 0.04019638897687678,
+      "step": 2535
+    }
+  ],
+  "logging_steps": 50,
+  "max_steps": 2252032,
+  "num_input_tokens_seen": 2401982,
+  "num_train_epochs": 4,
+  "save_steps": 507,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 0.0,
+  "train_batch_size": 2,
+  "trial_name": null,
+  "trial_params": null
+}

meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efd03f5456a4c15d2256ba2960a071c6dc2cb8a2ff8cf43b4f1331b6c112c442
+size 6737

meta-llama__llama-3.2-1b/grpo/run_manifest.json

@@ -6,10 +6,10 @@
   "model_name": "meta-llama/Llama-3.2-1B",
   "status": "running",
   "best_metric_name": "eval_cer_subset",
-  "best_metric_value": 0.012653280570618985,
-  "best_step": 1521,
-  "best_checkpoint_step": 1521,
-  "latest_checkpoint_step": 2028,
+  "best_metric_value": 0.011969319458693634,
+  "best_step": 2535,
+  "best_checkpoint_step": 2535,
+  "latest_checkpoint_step": 2535,
   "checkpoints": [
     {
       "step": 507,
@@ -34,10 +34,16 @@
       "epoch": 0.003602080254632261,
       "repo_path": "meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2028",
       "timestamp": "2026-04-09T01:23:03.985957+00:00"
+    },
+    {
+      "step": 2535,
+      "epoch": 0.004502600318290326,
+      "repo_path": "meta-llama__llama-3.2-1b/grpo/checkpoints/checkpoint-2535",
+      "timestamp": "2026-04-09T01:29:18.820055+00:00"
     }
   ],
   "epoch_artifacts": [],
   "resume_history": [],
   "created_at": "2026-04-09T00:54:21.973160+00:00",
-  "updated_at": "2026-04-09T01:23:03.985972+00:00"
+  "updated_at": "2026-04-09T01:29:18.820070+00:00"
 }