diff --git a/EVAL_REPORT.md b/EVAL_REPORT.md
new file mode 100644
index 0000000000000000000000000000000000000000..c8905e2536f2c62e8c2e69ac5530b22bd3095112
--- /dev/null
+++ b/EVAL_REPORT.md
@@ -0,0 +1,111 @@
+# Automated MNLP evaluation report
+
+- **Model repo:** [`cs-552-2026-the-transformers/math_model`](https://huggingface.co/cs-552-2026-the-transformers/math_model)
+- **Owner(s):** group **the-transformers**
+- **Generated at:** 2026-06-01T06:13:57+00:00 (UTC)
+- **Pipeline:** [mnlp-project-ci](https://github.com/eric11eca/mnlp-project-ci)
+
+_This PR is opened automatically by the course CI. It is **non-blocking** — you do not need to merge it. The next nightly run will refresh this file._
+
+## Summary
+
+| Benchmark | Accuracy | Status |
+|---|---:|---|
+| Math | 0.3800 | ok |
+| Knowledge | — | not run |
+| Multilingual | — | not run |
+| Safety | — | not run |
+
+## Sample completions
+
+_Prompts are intentionally omitted to avoid revealing benchmark contents. For multi-completion problems, only one completion is shown per sample._
+
+### Math
+
+**Correct** (1 shown)
+
+- **reference**: `$\frac{a}{2}$`
+- **overall** (1/8 completions correct)
+- **extracted** (✓): `\frac{a}{2}`
+- **completion**:
+
+ ```text
+
+ 1. **Identify Key Points and Relationships:**
+ - Given that $XYZ$ is an isosceles triangle with $XY = XZ$, the circumcenter $P$ lies on the perpendicular bisector of $YZ$.
+ - $Q$ is the midpoint of $YZ$, and $R$ is the reflection of $Q$ across $XZ$.
+ - $S$ is chosen such that $XQYS$ forms a rectangle.
+
+ 2. **Determine the Angles:**
+ - Since $XQYS$ is a rectangle, $\angle XQY = 90^\circ$.
+ - Because $Q$ is the midpoint of $YZ$, $QY = QZ$.
+ - Since $XQYS$ is a rectangle, $\angle XQS = 90^\circ$.
+ - Given $\angle YXZ = a$, we have $\angle QXZ = \frac{a}{2}$ because $Q$ is the midpoint of $YZ$.
+
+ 3. **Calculate $\angle QXR$:**
+ - Since $R$ is the reflection of $Q$ across $XZ$, $\angle QXR = \angle QXZ = \frac{a}{2}$.
+ - Therefore, $\angle QXR = \frac{a}{2}$.
+
+ 4. **Determine $\angle PRS$:**
+ - Since $P$ is the circumcenter of $\triangle XYZ$, $\angle PRS = \angle QXR$.
+ - Therefore, $\angle PRS = \frac{a}{2}$.
+
+ ### Conclusion:
+ \[
+
+
+ \boxed{\frac{a}{2}}
+ \]
+ ```
+
+**Incorrect** (1 shown)
+
+- **reference**: `756`
+- **overall** (0/8 completions correct)
+- **extracted** (✗): ``
+- **completion**:
+
+ ```text
+
+ To solve this problem, we need to find the squared distance between the centers of the three spherical balloons, given that the centers of the balloons lie on one side of a plane that cuts the balloons into three congruent circular cross-sections. We are given that the radii of the balloons are \(11\), \(13\), and \(19\), and the squared distance between the centers of the first and second
+
+ ... [truncated 821 chars] ...
+
+ 226. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 227. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 228. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 229. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 230. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 231. **Calculate the squared distance between the centers of the first and third balloons:**
+ Using the Pythagorean theorem again, we have:
+ \[
+ AC^2 = 11^2 + 19^2 = 121 + 361 = 482
+ \]
+
+ 232.
+ ```
diff --git a/README.md b/README.md
index 863e7d7f17ef0110a0719c682b315ea01f4e81a4..0009b8d26db78324045b065afa6b1cf95c073859 100644
--- a/README.md
+++ b/README.md
@@ -1,14 +1,15 @@
---
base_model: Qwen/Qwen3-1.7B
library_name: transformers
+model_name: exp6_sft_numinamath_dpo
tags:
- generated_from_trainer
+- dpo
- trl
-- sft
licence: license
---
-# math_model
+# Model Card for exp6_sft_numinamath_dpo
This model is a fine-tuned version of [Qwen/Qwen3-1.7B](https://huggingface.co/Qwen/Qwen3-1.7B).
It has been trained using [TRL](https://github.com/huggingface/trl).
@@ -26,10 +27,10 @@ print(output["generated_text"])
## Training procedure
-[
](https://wandb.ai/jdecim/MNLP/runs/2btc84l3)
+
-This model was trained with SFT.
+This model was trained with DPO, a method introduced in [Direct Preference Optimization: Your Language Model is Secretly a Reward Model](https://huggingface.co/papers/2305.18290).
### Framework versions
@@ -41,7 +42,18 @@ This model was trained with SFT.
## Citations
+Cite DPO as:
+```bibtex
+@inproceedings{rafailov2023direct,
+ title = {{Direct Preference Optimization: Your Language Model is Secretly a Reward Model}},
+ author = {Rafael Rafailov and Archit Sharma and Eric Mitchell and Christopher D. Manning and Stefano Ermon and Chelsea Finn},
+ year = 2023,
+ booktitle = {Advances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023},
+ url = {http://papers.nips.cc/paper_files/paper/2023/hash/a85b405ed65c6477a4fe8302b5e06ce7-Abstract-Conference.html},
+ editor = {Alice Oh and Tristan Naumann and Amir Globerson and Kate Saenko and Moritz Hardt and Sergey Levine},
+}
+```
Cite TRL as:
diff --git a/chat_template.jinja b/chat_template.jinja
index 5845311e60103eb4a734e3c4c8e69ba7348636a7..01be9b307daa2d425f7c168c9fb145a286e0afb4 100644
--- a/chat_template.jinja
+++ b/chat_template.jinja
@@ -10,13 +10,9 @@
{%- endfor %}
{{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
{%- else %}
- {#- BOXING INSTRUCTION injected so it is present even when the caller (e.g. CI)
- passes no system message. Targets our no-box / multi-box failure mode. #}
- {{- '<|im_start|>system\n' }}
{%- if messages[0].role == 'system' %}
- {{- messages[0].content + '\n\n' }}
+ {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
{%- endif %}
- {{- 'When solving a math problem, work through it step by step and write your final answer inside a \boxed{...}.<|im_end|>\n' }}
{%- endif %}
{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
{%- for message in messages[::-1] %}
diff --git a/checkpoint-1011/chat_template.jinja b/checkpoint-1011/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-1011/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-1011/config.json b/checkpoint-1011/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-1011/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-1011/generation_config.json b/checkpoint-1011/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-1011/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-1011/model.safetensors b/checkpoint-1011/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cc885a501a0fffef97b843b5248e3bb7cfe46172
--- /dev/null
+++ b/checkpoint-1011/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3af4557502ae7c7853ed3b684aa50cfd00317aafc237f2dda688075bb3dc81c1
+size 3441185608
diff --git a/checkpoint-1011/optimizer.pt b/checkpoint-1011/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..d00b4ce7da17dce3a0fb31d9c369e0cafbfc68a3
--- /dev/null
+++ b/checkpoint-1011/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:016daefe3fcafd2822ed799155138cb8e476fe309f9070f5157970a18bcc6d4c
+size 6882572207
diff --git a/checkpoint-1011/rng_state.pth b/checkpoint-1011/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..bd3c190a62ba12f21428f29e0f4bde711034a75b
--- /dev/null
+++ b/checkpoint-1011/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4a9f217e852f439efa6bd32fde98d6867f11aa6ea13ddc021ba10af6a0b0934
+size 14645
diff --git a/checkpoint-1011/scheduler.pt b/checkpoint-1011/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..edf0d85a22eaf7a475ed68221fba576eaa55f682
--- /dev/null
+++ b/checkpoint-1011/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ee946d348a6273229e02f92d4563cb2ba7ac75075130a9d6c5c0eb31d584ddc
+size 1465
diff --git a/checkpoint-1011/tokenizer.json b/checkpoint-1011/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-1011/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-1011/tokenizer_config.json b/checkpoint-1011/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-1011/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-1011/trainer_state.json b/checkpoint-1011/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..9becb660138c318878d26d4efdaa5944723c7ae4
--- /dev/null
+++ b/checkpoint-1011/trainer_state.json
@@ -0,0 +1,1044 @@
+{
+ "best_global_step": null,
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 3.0,
+ "eval_steps": 500,
+ "global_step": 1011,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": NaN,
+ "epoch": 0.029717682020802376,
+ "grad_norm": NaN,
+ "learning_rate": 1.8000000000000001e-06,
+ "loss": 1005514863411.2,
+ "mean_token_accuracy": 0.07770282882993343,
+ "num_tokens": 97764.0,
+ "step": 10
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.05943536404160475,
+ "grad_norm": NaN,
+ "learning_rate": 3.8000000000000005e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007392044077278115,
+ "num_tokens": 190372.0,
+ "step": 20
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.08915304606240713,
+ "grad_norm": NaN,
+ "learning_rate": 5.8e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000383764869911829,
+ "num_tokens": 280928.0,
+ "step": 30
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.1188707280832095,
+ "grad_norm": NaN,
+ "learning_rate": 7.800000000000002e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003212536917999387,
+ "num_tokens": 380644.0,
+ "step": 40
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.1485884101040119,
+ "grad_norm": NaN,
+ "learning_rate": 9.800000000000001e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003853117574180942,
+ "num_tokens": 475467.0,
+ "step": 50
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.17830609212481427,
+ "grad_norm": NaN,
+ "learning_rate": 1.18e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003134153150313068,
+ "num_tokens": 572411.0,
+ "step": 60
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.20802377414561665,
+ "grad_norm": NaN,
+ "learning_rate": 1.38e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011001345737895463,
+ "num_tokens": 661517.0,
+ "step": 70
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.237741456166419,
+ "grad_norm": NaN,
+ "learning_rate": 1.58e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012231820757733658,
+ "num_tokens": 752142.0,
+ "step": 80
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.2674591381872214,
+ "grad_norm": NaN,
+ "learning_rate": 1.7800000000000002e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011551132018212229,
+ "num_tokens": 850399.0,
+ "step": 90
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.2971768202080238,
+ "grad_norm": NaN,
+ "learning_rate": 1.98e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00047943509634933433,
+ "num_tokens": 954759.0,
+ "step": 100
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.32689450222882616,
+ "grad_norm": NaN,
+ "learning_rate": 1.9802414928649837e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008438902725174558,
+ "num_tokens": 1050439.0,
+ "step": 110
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.35661218424962854,
+ "grad_norm": NaN,
+ "learning_rate": 1.958287596048299e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009903057703922967,
+ "num_tokens": 1135089.0,
+ "step": 120
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.3863298662704309,
+ "grad_norm": NaN,
+ "learning_rate": 1.9363336992316137e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00040032515971688554,
+ "num_tokens": 1228149.0,
+ "step": 130
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.4160475482912333,
+ "grad_norm": NaN,
+ "learning_rate": 1.9143798024149288e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0010247966441966128,
+ "num_tokens": 1331672.0,
+ "step": 140
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.4457652303120357,
+ "grad_norm": NaN,
+ "learning_rate": 1.892425905598244e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000843161555530969,
+ "num_tokens": 1421314.0,
+ "step": 150
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.475482912332838,
+ "grad_norm": NaN,
+ "learning_rate": 1.8704720087815587e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007889680869993754,
+ "num_tokens": 1511701.0,
+ "step": 160
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5052005943536404,
+ "grad_norm": NaN,
+ "learning_rate": 1.848518111964874e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007493881763366517,
+ "num_tokens": 1605063.0,
+ "step": 170
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5349182763744428,
+ "grad_norm": NaN,
+ "learning_rate": 1.826564215148189e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000567350817436818,
+ "num_tokens": 1703107.0,
+ "step": 180
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5646359583952452,
+ "grad_norm": NaN,
+ "learning_rate": 1.804610318331504e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00027060438151238486,
+ "num_tokens": 1796279.0,
+ "step": 190
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5943536404160475,
+ "grad_norm": NaN,
+ "learning_rate": 1.782656421514819e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008226798396208324,
+ "num_tokens": 1893647.0,
+ "step": 200
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6240713224368499,
+ "grad_norm": NaN,
+ "learning_rate": 1.760702524698134e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007338574505411088,
+ "num_tokens": 1982325.0,
+ "step": 210
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6537890044576523,
+ "grad_norm": NaN,
+ "learning_rate": 1.7387486278814492e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004891513846814633,
+ "num_tokens": 2071869.0,
+ "step": 220
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6835066864784547,
+ "grad_norm": NaN,
+ "learning_rate": 1.716794731064764e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00038509477453771976,
+ "num_tokens": 2168370.0,
+ "step": 230
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7132243684992571,
+ "grad_norm": NaN,
+ "learning_rate": 1.694840834248079e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009255842014681548,
+ "num_tokens": 2263211.0,
+ "step": 240
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7429420505200595,
+ "grad_norm": NaN,
+ "learning_rate": 1.6728869374313943e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011857455348945223,
+ "num_tokens": 2352425.0,
+ "step": 250
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7726597325408618,
+ "grad_norm": NaN,
+ "learning_rate": 1.6509330406147094e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008204675337765366,
+ "num_tokens": 2445217.0,
+ "step": 260
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8023774145616642,
+ "grad_norm": NaN,
+ "learning_rate": 1.6289791437980242e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0013553448639868292,
+ "num_tokens": 2537677.0,
+ "step": 270
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8320950965824666,
+ "grad_norm": NaN,
+ "learning_rate": 1.6070252469813393e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011630529144895263,
+ "num_tokens": 2631646.0,
+ "step": 280
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.861812778603269,
+ "grad_norm": NaN,
+ "learning_rate": 1.5850713501646545e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009409735575900413,
+ "num_tokens": 2722632.0,
+ "step": 290
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8915304606240714,
+ "grad_norm": NaN,
+ "learning_rate": 1.5631174533479693e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006484775040007662,
+ "num_tokens": 2819185.0,
+ "step": 300
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.9212481426448736,
+ "grad_norm": NaN,
+ "learning_rate": 1.5411635565312844e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011682231881422922,
+ "num_tokens": 2910483.0,
+ "step": 310
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.950965824665676,
+ "grad_norm": NaN,
+ "learning_rate": 1.5192096597145995e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004708930806373246,
+ "num_tokens": 2997057.0,
+ "step": 320
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.9806835066864784,
+ "grad_norm": NaN,
+ "learning_rate": 1.4972557628979145e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011955970534472727,
+ "num_tokens": 3091284.0,
+ "step": 330
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0089153046062407,
+ "grad_norm": NaN,
+ "learning_rate": 1.4753018660812295e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006843029964090276,
+ "num_tokens": 3177800.0,
+ "step": 340
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0386329866270432,
+ "grad_norm": NaN,
+ "learning_rate": 1.4533479692645444e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005406756106822285,
+ "num_tokens": 3271989.0,
+ "step": 350
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0683506686478454,
+ "grad_norm": NaN,
+ "learning_rate": 1.4313940724478596e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007009803419350646,
+ "num_tokens": 3365082.0,
+ "step": 360
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.098068350668648,
+ "grad_norm": NaN,
+ "learning_rate": 1.4094401756311746e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001003894711175235,
+ "num_tokens": 3465865.0,
+ "step": 370
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.1277860326894502,
+ "grad_norm": NaN,
+ "learning_rate": 1.3874862788144897e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004136150790145621,
+ "num_tokens": 3564998.0,
+ "step": 380
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.1575037147102525,
+ "grad_norm": NaN,
+ "learning_rate": 1.3655323819978048e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000998112405068241,
+ "num_tokens": 3661508.0,
+ "step": 390
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.187221396731055,
+ "grad_norm": NaN,
+ "learning_rate": 1.3435784851811198e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009621107514249161,
+ "num_tokens": 3755395.0,
+ "step": 400
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.2169390787518575,
+ "grad_norm": NaN,
+ "learning_rate": 1.3216245883644348e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004976072828867473,
+ "num_tokens": 3842994.0,
+ "step": 410
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.2466567607726597,
+ "grad_norm": NaN,
+ "learning_rate": 1.2996706915477499e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003434549740632065,
+ "num_tokens": 3934618.0,
+ "step": 420
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.276374442793462,
+ "grad_norm": NaN,
+ "learning_rate": 1.2777167947310649e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00026364591685705817,
+ "num_tokens": 4026711.0,
+ "step": 430
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3060921248142645,
+ "grad_norm": NaN,
+ "learning_rate": 1.2557628979143798e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006349188319290988,
+ "num_tokens": 4114609.0,
+ "step": 440
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3358098068350668,
+ "grad_norm": NaN,
+ "learning_rate": 1.2338090010976948e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004265022376785055,
+ "num_tokens": 4203953.0,
+ "step": 450
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3655274888558693,
+ "grad_norm": NaN,
+ "learning_rate": 1.2118551042810101e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012633154692593962,
+ "num_tokens": 4298319.0,
+ "step": 460
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3952451708766715,
+ "grad_norm": NaN,
+ "learning_rate": 1.189901207464325e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011256658195634372,
+ "num_tokens": 4396020.0,
+ "step": 470
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.424962852897474,
+ "grad_norm": NaN,
+ "learning_rate": 1.16794731064764e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0013131745137798135,
+ "num_tokens": 4491203.0,
+ "step": 480
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.4546805349182763,
+ "grad_norm": NaN,
+ "learning_rate": 1.1459934138309552e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012716218378045596,
+ "num_tokens": 4582199.0,
+ "step": 490
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.4843982169390788,
+ "grad_norm": NaN,
+ "learning_rate": 1.1240395170142701e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0014869970778818241,
+ "num_tokens": 4672589.0,
+ "step": 500
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.514115898959881,
+ "grad_norm": NaN,
+ "learning_rate": 1.1020856201975851e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00046981508567114363,
+ "num_tokens": 4767693.0,
+ "step": 510
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.5438335809806834,
+ "grad_norm": NaN,
+ "learning_rate": 1.0801317233809e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000601445968641201,
+ "num_tokens": 4860371.0,
+ "step": 520
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.5735512630014858,
+ "grad_norm": NaN,
+ "learning_rate": 1.0581778265642154e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012073652920662425,
+ "num_tokens": 4950622.0,
+ "step": 530
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6032689450222883,
+ "grad_norm": NaN,
+ "learning_rate": 1.0362239297475303e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00028707647215924225,
+ "num_tokens": 5044878.0,
+ "step": 540
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6329866270430906,
+ "grad_norm": NaN,
+ "learning_rate": 1.0142700329308453e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004101752907445189,
+ "num_tokens": 5140938.0,
+ "step": 550
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6627043090638929,
+ "grad_norm": NaN,
+ "learning_rate": 9.923161361141603e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008072987395280507,
+ "num_tokens": 5232852.0,
+ "step": 560
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6924219910846954,
+ "grad_norm": NaN,
+ "learning_rate": 9.703622392974754e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012222853780258446,
+ "num_tokens": 5321835.0,
+ "step": 570
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7221396731054979,
+ "grad_norm": NaN,
+ "learning_rate": 9.484083424807904e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001064640874392353,
+ "num_tokens": 5416464.0,
+ "step": 580
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7518573551263001,
+ "grad_norm": NaN,
+ "learning_rate": 9.264544456641055e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009612867230316624,
+ "num_tokens": 5515169.0,
+ "step": 590
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7815750371471024,
+ "grad_norm": NaN,
+ "learning_rate": 9.045005488474205e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008774219051701948,
+ "num_tokens": 5611971.0,
+ "step": 600
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.811292719167905,
+ "grad_norm": NaN,
+ "learning_rate": 8.825466520307355e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007863394406740553,
+ "num_tokens": 5704255.0,
+ "step": 610
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.8410104011887074,
+ "grad_norm": NaN,
+ "learning_rate": 8.605927552140506e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000578320066415472,
+ "num_tokens": 5800440.0,
+ "step": 620
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.8707280832095097,
+ "grad_norm": NaN,
+ "learning_rate": 8.386388583973656e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004275172817870043,
+ "num_tokens": 5890898.0,
+ "step": 630
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.900445765230312,
+ "grad_norm": NaN,
+ "learning_rate": 8.166849615806807e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001425291335908696,
+ "num_tokens": 5987228.0,
+ "step": 640
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.9301634472511144,
+ "grad_norm": NaN,
+ "learning_rate": 7.947310647639957e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006059414023184218,
+ "num_tokens": 6076579.0,
+ "step": 650
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.959881129271917,
+ "grad_norm": NaN,
+ "learning_rate": 7.727771679473108e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008722793892957271,
+ "num_tokens": 6171013.0,
+ "step": 660
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.9895988112927192,
+ "grad_norm": NaN,
+ "learning_rate": 7.508232711306257e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007748565993097145,
+ "num_tokens": 6265432.0,
+ "step": 670
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.0178306092124814,
+ "grad_norm": NaN,
+ "learning_rate": 7.288693743139408e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007497724390735751,
+ "num_tokens": 6349548.0,
+ "step": 680
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.0475482912332836,
+ "grad_norm": NaN,
+ "learning_rate": 7.069154774972559e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006858149892650545,
+ "num_tokens": 6440481.0,
+ "step": 690
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.0772659732540864,
+ "grad_norm": NaN,
+ "learning_rate": 6.849615806805708e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008518316331901588,
+ "num_tokens": 6533227.0,
+ "step": 700
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.1069836552748886,
+ "grad_norm": NaN,
+ "learning_rate": 6.630076838638859e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006546900600369554,
+ "num_tokens": 6631305.0,
+ "step": 710
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.136701337295691,
+ "grad_norm": NaN,
+ "learning_rate": 6.410537870472009e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006311897363048047,
+ "num_tokens": 6719619.0,
+ "step": 720
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.166419019316493,
+ "grad_norm": NaN,
+ "learning_rate": 6.19099890230516e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008787176091573201,
+ "num_tokens": 6808489.0,
+ "step": 730
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.196136701337296,
+ "grad_norm": NaN,
+ "learning_rate": 5.9714599341383095e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0010024115661508403,
+ "num_tokens": 6904544.0,
+ "step": 740
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.225854383358098,
+ "grad_norm": NaN,
+ "learning_rate": 5.75192096597146e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006462240009568632,
+ "num_tokens": 6998909.0,
+ "step": 750
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.2555720653789004,
+ "grad_norm": NaN,
+ "learning_rate": 5.532381997804611e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005598854688287247,
+ "num_tokens": 7093305.0,
+ "step": 760
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.2852897473997027,
+ "grad_norm": NaN,
+ "learning_rate": 5.312843029637761e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00036793685212614945,
+ "num_tokens": 7184216.0,
+ "step": 770
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.315007429420505,
+ "grad_norm": NaN,
+ "learning_rate": 5.093304061470912e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000779397370206425,
+ "num_tokens": 7273139.0,
+ "step": 780
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.3447251114413077,
+ "grad_norm": NaN,
+ "learning_rate": 4.873765093304062e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006776292997528799,
+ "num_tokens": 7375877.0,
+ "step": 790
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.37444279346211,
+ "grad_norm": NaN,
+ "learning_rate": 4.654226125137213e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0013253977747808676,
+ "num_tokens": 7471802.0,
+ "step": 800
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.4041604754829122,
+ "grad_norm": NaN,
+ "learning_rate": 4.434687156970362e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007797809317708016,
+ "num_tokens": 7568137.0,
+ "step": 810
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.433878157503715,
+ "grad_norm": NaN,
+ "learning_rate": 4.215148188803513e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005893595473025926,
+ "num_tokens": 7662880.0,
+ "step": 820
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.463595839524517,
+ "grad_norm": NaN,
+ "learning_rate": 3.995609220636663e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007197739760158584,
+ "num_tokens": 7755580.0,
+ "step": 830
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.4933135215453195,
+ "grad_norm": NaN,
+ "learning_rate": 3.776070252469814e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000664884723664727,
+ "num_tokens": 7854588.0,
+ "step": 840
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.5230312035661218,
+ "grad_norm": NaN,
+ "learning_rate": 3.5565312843029644e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000588176279416075,
+ "num_tokens": 7950417.0,
+ "step": 850
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.552748885586924,
+ "grad_norm": NaN,
+ "learning_rate": 3.3369923161361144e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007997245862497948,
+ "num_tokens": 8043286.0,
+ "step": 860
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.5824665676077267,
+ "grad_norm": NaN,
+ "learning_rate": 3.117453347969265e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005276548559777439,
+ "num_tokens": 8138513.0,
+ "step": 870
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.612184249628529,
+ "grad_norm": NaN,
+ "learning_rate": 2.897914379802415e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00023305922804865985,
+ "num_tokens": 8228539.0,
+ "step": 880
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.6419019316493313,
+ "grad_norm": NaN,
+ "learning_rate": 2.6783754116355656e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005359479633625597,
+ "num_tokens": 8320083.0,
+ "step": 890
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.6716196136701336,
+ "grad_norm": NaN,
+ "learning_rate": 2.458836443468716e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006636981313931755,
+ "num_tokens": 8416446.0,
+ "step": 900
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.7013372956909363,
+ "grad_norm": NaN,
+ "learning_rate": 2.239297475301866e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.002058879991818685,
+ "num_tokens": 8513551.0,
+ "step": 910
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.7310549777117386,
+ "grad_norm": NaN,
+ "learning_rate": 2.0197585071350167e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000950983911025105,
+ "num_tokens": 8603909.0,
+ "step": 920
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.760772659732541,
+ "grad_norm": NaN,
+ "learning_rate": 1.800219538968167e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0014912645376170986,
+ "num_tokens": 8695081.0,
+ "step": 930
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.790490341753343,
+ "grad_norm": NaN,
+ "learning_rate": 1.5806805708013173e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001528770748700481,
+ "num_tokens": 8790760.0,
+ "step": 940
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.8202080237741454,
+ "grad_norm": NaN,
+ "learning_rate": 1.3611416026344676e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012167295863036998,
+ "num_tokens": 8879103.0,
+ "step": 950
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.849925705794948,
+ "grad_norm": NaN,
+ "learning_rate": 1.1416026344676181e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007455661499989219,
+ "num_tokens": 8972354.0,
+ "step": 960
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.8796433878157504,
+ "grad_norm": NaN,
+ "learning_rate": 9.220636663007684e-07,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003935990913305432,
+ "num_tokens": 9069381.0,
+ "step": 970
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.9093610698365526,
+ "grad_norm": NaN,
+ "learning_rate": 7.025246981339188e-07,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00037132185607333665,
+ "num_tokens": 9164901.0,
+ "step": 980
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.9390787518573553,
+ "grad_norm": NaN,
+ "learning_rate": 4.829857299670692e-07,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00036687121028080585,
+ "num_tokens": 9258811.0,
+ "step": 990
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.9687964338781576,
+ "grad_norm": NaN,
+ "learning_rate": 2.6344676180021955e-07,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008805370052868966,
+ "num_tokens": 9351174.0,
+ "step": 1000
+ },
+ {
+ "entropy": NaN,
+ "epoch": 2.99851411589896,
+ "grad_norm": NaN,
+ "learning_rate": 4.390779363336993e-08,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007362255608313717,
+ "num_tokens": 9440239.0,
+ "step": 1010
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 1011,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 3,
+ "save_steps": 500,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": true
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 1.0120089572705894e+17,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-1011/training_args.bin b/checkpoint-1011/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..737a87caa4fc2f704a1fe56a508ca4d335ca783d
--- /dev/null
+++ b/checkpoint-1011/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae0f7084660199ed7fd16027c0c6af5e726790f32a022b019f1277bec4e01142
+size 5713
diff --git a/checkpoint-125/chat_template.jinja b/checkpoint-125/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-125/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-125/config.json b/checkpoint-125/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-125/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-125/generation_config.json b/checkpoint-125/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-125/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-125/model.safetensors b/checkpoint-125/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..d2e0d2f571547f58479dee02aa685fc8990a4284
--- /dev/null
+++ b/checkpoint-125/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7074b51b321535051b77436f30720111854403259c583a273488e061c0d23072
+size 3441185608
diff --git a/checkpoint-125/optimizer.pt b/checkpoint-125/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c18832d9f63e5dc9e826ce8b7a6fa36249242ab6
--- /dev/null
+++ b/checkpoint-125/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b569f0710425a088cc5cd95c2def7bbb89539cdd1c3b41312c8d8fc136d2e186
+size 6882572207
diff --git a/checkpoint-125/rng_state.pth b/checkpoint-125/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..5a8f17addf8aaee350625e9847751582bc9f86dd
--- /dev/null
+++ b/checkpoint-125/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:61c19bab1174704a4a4441475683bf1270277af15d2e2c95e964789128e482c4
+size 14645
diff --git a/checkpoint-125/scheduler.pt b/checkpoint-125/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..c0f9744d60b0805d01c098d6c1052451f23dd980
--- /dev/null
+++ b/checkpoint-125/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:770e85488b2d6f2b41a590e6c2bb599f3a174081e14b8c48d3638681b4a06597
+size 1465
diff --git a/checkpoint-125/tokenizer.json b/checkpoint-125/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-125/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-125/tokenizer_config.json b/checkpoint-125/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-125/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-125/trainer_state.json b/checkpoint-125/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..e2e8da443bc02a76e800bb696414257eb8fe5e6c
--- /dev/null
+++ b/checkpoint-125/trainer_state.json
@@ -0,0 +1,154 @@
+{
+ "best_global_step": null,
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 1.0,
+ "eval_steps": 500,
+ "global_step": 125,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": 0.2860680317506194,
+ "epoch": 0.08,
+ "grad_norm": 7.46875,
+ "learning_rate": 1.5000000000000002e-05,
+ "loss": 0.7724831104278564,
+ "mean_token_accuracy": 0.8434554561972618,
+ "num_tokens": 94518.0,
+ "step": 10
+ },
+ {
+ "entropy": 0.3743372498080134,
+ "epoch": 0.16,
+ "grad_norm": 2.6875,
+ "learning_rate": 1.8761061946902657e-05,
+ "loss": 0.4372231483459473,
+ "mean_token_accuracy": 0.8732902690768242,
+ "num_tokens": 187445.0,
+ "step": 20
+ },
+ {
+ "entropy": 0.3575092004612088,
+ "epoch": 0.24,
+ "grad_norm": 2.625,
+ "learning_rate": 1.6991150442477876e-05,
+ "loss": 0.3412153959274292,
+ "mean_token_accuracy": 0.8936023108661175,
+ "num_tokens": 280610.0,
+ "step": 30
+ },
+ {
+ "entropy": 0.31294157058000566,
+ "epoch": 0.32,
+ "grad_norm": 2.421875,
+ "learning_rate": 1.5221238938053098e-05,
+ "loss": 0.3175334453582764,
+ "mean_token_accuracy": 0.8971926532685757,
+ "num_tokens": 379577.0,
+ "step": 40
+ },
+ {
+ "entropy": 0.30759106744080783,
+ "epoch": 0.4,
+ "grad_norm": 1.7421875,
+ "learning_rate": 1.345132743362832e-05,
+ "loss": 0.32603118419647215,
+ "mean_token_accuracy": 0.8967837899923324,
+ "num_tokens": 477577.0,
+ "step": 50
+ },
+ {
+ "entropy": 0.31574366632848977,
+ "epoch": 0.48,
+ "grad_norm": 1.9921875,
+ "learning_rate": 1.1681415929203541e-05,
+ "loss": 0.323792552947998,
+ "mean_token_accuracy": 0.8988608829677105,
+ "num_tokens": 573360.0,
+ "step": 60
+ },
+ {
+ "entropy": 0.3065364420413971,
+ "epoch": 0.56,
+ "grad_norm": 1.6171875,
+ "learning_rate": 9.911504424778762e-06,
+ "loss": 0.32266840934753416,
+ "mean_token_accuracy": 0.8985385209321975,
+ "num_tokens": 665761.0,
+ "step": 70
+ },
+ {
+ "entropy": 0.27206654362380506,
+ "epoch": 0.64,
+ "grad_norm": 2.1875,
+ "learning_rate": 8.141592920353984e-06,
+ "loss": 0.2854612350463867,
+ "mean_token_accuracy": 0.9085792370140553,
+ "num_tokens": 756921.0,
+ "step": 80
+ },
+ {
+ "entropy": 0.28962985165417193,
+ "epoch": 0.72,
+ "grad_norm": 1.8125,
+ "learning_rate": 6.371681415929204e-06,
+ "loss": 0.29041600227355957,
+ "mean_token_accuracy": 0.9059965394437313,
+ "num_tokens": 845171.0,
+ "step": 90
+ },
+ {
+ "entropy": 0.2753044681623578,
+ "epoch": 0.8,
+ "grad_norm": 2.09375,
+ "learning_rate": 4.6017699115044254e-06,
+ "loss": 0.2797576427459717,
+ "mean_token_accuracy": 0.9101039692759514,
+ "num_tokens": 939550.0,
+ "step": 100
+ },
+ {
+ "entropy": 0.2808386910706758,
+ "epoch": 0.88,
+ "grad_norm": 1.5859375,
+ "learning_rate": 2.831858407079646e-06,
+ "loss": 0.2893812894821167,
+ "mean_token_accuracy": 0.9063553236424923,
+ "num_tokens": 1041441.0,
+ "step": 110
+ },
+ {
+ "entropy": 0.2944059643894434,
+ "epoch": 0.96,
+ "grad_norm": 1.6875,
+ "learning_rate": 1.0619469026548673e-06,
+ "loss": 0.2996867895126343,
+ "mean_token_accuracy": 0.9011975660920143,
+ "num_tokens": 1132794.0,
+ "step": 120
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 125,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 1,
+ "save_steps": 500,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": true
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 1.2589656731283456e+16,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-125/training_args.bin b/checkpoint-125/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..0b1ceadcc46a4dbcb04d34405de43a2667755f2a
--- /dev/null
+++ b/checkpoint-125/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4e0921a533d54992c732b7cf0c8a29ef97e50bbd333d4e4c07ae7183a1feb0f
+size 5713
diff --git a/checkpoint-500/chat_template.jinja b/checkpoint-500/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-500/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-500/config.json b/checkpoint-500/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-500/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-500/generation_config.json b/checkpoint-500/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-500/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-500/model.safetensors b/checkpoint-500/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..aad24a30496616243a28537371ee273f78ce5980
--- /dev/null
+++ b/checkpoint-500/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:079565c41b7390c29162b2177ece2956e65a2c94398b94f8fbdcc9aaea1e5987
+size 3441185608
diff --git a/checkpoint-500/optimizer.pt b/checkpoint-500/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..90c55c866e33ae7b8a57d59da6cb54957b236309
--- /dev/null
+++ b/checkpoint-500/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e11662089759b5ab33db1036a43acc423329b165030883aa32639c9ab33f035f
+size 6882572207
diff --git a/checkpoint-500/rng_state.pth b/checkpoint-500/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..3105ae629835438feca5d75cb928a0254192bb7f
--- /dev/null
+++ b/checkpoint-500/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f196323d7423b60f8e4ceb7dbf8715ee326c0d068e5ff164f13c63b279b9f1a0
+size 14645
diff --git a/checkpoint-500/scheduler.pt b/checkpoint-500/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..3cbc1d4f12d934fbe27c91c192400d02b27d2e00
--- /dev/null
+++ b/checkpoint-500/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:96ae61b7327de513c572cbdd449d40136fa2f4fac8357d34b3d1a844dc5168d3
+size 1465
diff --git a/checkpoint-500/tokenizer.json b/checkpoint-500/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-500/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-500/tokenizer_config.json b/checkpoint-500/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-500/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-500/trainer_state.json b/checkpoint-500/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..511dccfadd29552a3ad6e861310aa24e146a179e
--- /dev/null
+++ b/checkpoint-500/trainer_state.json
@@ -0,0 +1,589 @@
+{
+ "best_global_step": 500,
+ "best_metric": 0.2841346263885498,
+ "best_model_checkpoint": "/scratch/checkpoints/math_sft_v2/checkpoint-500",
+ "epoch": 1.6507018992568125,
+ "eval_steps": 100,
+ "global_step": 500,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": 0.26208699569106103,
+ "epoch": 0.03303055326176713,
+ "grad_norm": 13.3125,
+ "learning_rate": 1.5e-06,
+ "loss": 0.8376299858093261,
+ "mean_token_accuracy": 0.8421687923371792,
+ "num_tokens": 98215.0,
+ "step": 10
+ },
+ {
+ "entropy": 0.2689803125336766,
+ "epoch": 0.06606110652353427,
+ "grad_norm": 11.5625,
+ "learning_rate": 3.1666666666666667e-06,
+ "loss": 0.853145980834961,
+ "mean_token_accuracy": 0.8399689704179764,
+ "num_tokens": 191821.0,
+ "step": 20
+ },
+ {
+ "entropy": 0.28777916859835384,
+ "epoch": 0.0990916597853014,
+ "grad_norm": 12.375,
+ "learning_rate": 4.833333333333333e-06,
+ "loss": 0.7679028034210205,
+ "mean_token_accuracy": 0.8431851923465729,
+ "num_tokens": 285491.0,
+ "step": 30
+ },
+ {
+ "entropy": 0.3033197607845068,
+ "epoch": 0.13212221304706853,
+ "grad_norm": 5.375,
+ "learning_rate": 6.5000000000000004e-06,
+ "loss": 0.6137509822845459,
+ "mean_token_accuracy": 0.8549904882907867,
+ "num_tokens": 381273.0,
+ "step": 40
+ },
+ {
+ "entropy": 0.34702248610556125,
+ "epoch": 0.16515276630883569,
+ "grad_norm": 4.25,
+ "learning_rate": 8.166666666666668e-06,
+ "loss": 0.5155063152313233,
+ "mean_token_accuracy": 0.8607264116406441,
+ "num_tokens": 476481.0,
+ "step": 50
+ },
+ {
+ "entropy": 0.35821260381489994,
+ "epoch": 0.1981833195706028,
+ "grad_norm": 2.890625,
+ "learning_rate": 9.833333333333333e-06,
+ "loss": 0.42441272735595703,
+ "mean_token_accuracy": 0.8774531938135623,
+ "num_tokens": 570979.0,
+ "step": 60
+ },
+ {
+ "entropy": 0.33337059412151576,
+ "epoch": 0.23121387283236994,
+ "grad_norm": 3.15625,
+ "learning_rate": 9.835164835164835e-06,
+ "loss": 0.34101109504699706,
+ "mean_token_accuracy": 0.8939957685768605,
+ "num_tokens": 664789.0,
+ "step": 70
+ },
+ {
+ "entropy": 0.30876336991786957,
+ "epoch": 0.26424442609413706,
+ "grad_norm": 1.8515625,
+ "learning_rate": 9.652014652014653e-06,
+ "loss": 0.32317628860473635,
+ "mean_token_accuracy": 0.8987478196620942,
+ "num_tokens": 764246.0,
+ "step": 80
+ },
+ {
+ "entropy": 0.30824985895305873,
+ "epoch": 0.2972749793559042,
+ "grad_norm": 2.15625,
+ "learning_rate": 9.46886446886447e-06,
+ "loss": 0.31807353496551516,
+ "mean_token_accuracy": 0.896314675360918,
+ "num_tokens": 855689.0,
+ "step": 90
+ },
+ {
+ "entropy": 0.31522620394825934,
+ "epoch": 0.33030553261767137,
+ "grad_norm": 1.8828125,
+ "learning_rate": 9.285714285714288e-06,
+ "loss": 0.338785982131958,
+ "mean_token_accuracy": 0.89400485008955,
+ "num_tokens": 949366.0,
+ "step": 100
+ },
+ {
+ "epoch": 0.33030553261767137,
+ "eval_entropy": 0.3022541454544774,
+ "eval_loss": 0.3146008253097534,
+ "eval_mean_token_accuracy": 0.8988911988558592,
+ "eval_num_tokens": 949366.0,
+ "eval_runtime": 17.914,
+ "eval_samples_per_second": 30.088,
+ "eval_steps_per_second": 15.072,
+ "step": 100
+ },
+ {
+ "entropy": 0.2985268572345376,
+ "epoch": 0.3633360858794385,
+ "grad_norm": 2.171875,
+ "learning_rate": 9.102564102564104e-06,
+ "loss": 0.31417596340179443,
+ "mean_token_accuracy": 0.89812126532197,
+ "num_tokens": 1040854.0,
+ "step": 110
+ },
+ {
+ "entropy": 0.3040375892072916,
+ "epoch": 0.3963666391412056,
+ "grad_norm": 1.9609375,
+ "learning_rate": 8.91941391941392e-06,
+ "loss": 0.32050628662109376,
+ "mean_token_accuracy": 0.8970061101019382,
+ "num_tokens": 1132424.0,
+ "step": 120
+ },
+ {
+ "entropy": 0.3125588171184063,
+ "epoch": 0.42939719240297275,
+ "grad_norm": 1.8203125,
+ "learning_rate": 8.736263736263737e-06,
+ "loss": 0.3176266193389893,
+ "mean_token_accuracy": 0.8971590586006641,
+ "num_tokens": 1221899.0,
+ "step": 130
+ },
+ {
+ "entropy": 0.312872576713562,
+ "epoch": 0.4624277456647399,
+ "grad_norm": 1.78125,
+ "learning_rate": 8.553113553113553e-06,
+ "loss": 0.3343791484832764,
+ "mean_token_accuracy": 0.8940604299306869,
+ "num_tokens": 1315052.0,
+ "step": 140
+ },
+ {
+ "entropy": 0.27830475065857174,
+ "epoch": 0.495458298926507,
+ "grad_norm": 2.0,
+ "learning_rate": 8.369963369963371e-06,
+ "loss": 0.28718521595001223,
+ "mean_token_accuracy": 0.9051955968141556,
+ "num_tokens": 1401947.0,
+ "step": 150
+ },
+ {
+ "entropy": 0.3005422970280051,
+ "epoch": 0.5284888521882741,
+ "grad_norm": 2.34375,
+ "learning_rate": 8.186813186813188e-06,
+ "loss": 0.31569790840148926,
+ "mean_token_accuracy": 0.8999212823808194,
+ "num_tokens": 1494107.0,
+ "step": 160
+ },
+ {
+ "entropy": 0.31200682315975425,
+ "epoch": 0.5615194054500413,
+ "grad_norm": 2.015625,
+ "learning_rate": 8.003663003663006e-06,
+ "loss": 0.32325453758239747,
+ "mean_token_accuracy": 0.8952137432992459,
+ "num_tokens": 1586474.0,
+ "step": 170
+ },
+ {
+ "entropy": 0.2929861258715391,
+ "epoch": 0.5945499587118084,
+ "grad_norm": 2.09375,
+ "learning_rate": 7.820512820512822e-06,
+ "loss": 0.29693677425384524,
+ "mean_token_accuracy": 0.9031261466443539,
+ "num_tokens": 1678162.0,
+ "step": 180
+ },
+ {
+ "entropy": 0.30670158881694076,
+ "epoch": 0.6275805119735756,
+ "grad_norm": 1.828125,
+ "learning_rate": 7.637362637362638e-06,
+ "loss": 0.31567308902740476,
+ "mean_token_accuracy": 0.8983482263982296,
+ "num_tokens": 1781401.0,
+ "step": 190
+ },
+ {
+ "entropy": 0.2866023685783148,
+ "epoch": 0.6606110652353427,
+ "grad_norm": 1.6953125,
+ "learning_rate": 7.454212454212455e-06,
+ "loss": 0.29333360195159913,
+ "mean_token_accuracy": 0.904055442661047,
+ "num_tokens": 1868038.0,
+ "step": 200
+ },
+ {
+ "epoch": 0.6606110652353427,
+ "eval_entropy": 0.2921138784951634,
+ "eval_loss": 0.29292193055152893,
+ "eval_mean_token_accuracy": 0.904505420172656,
+ "eval_num_tokens": 1868038.0,
+ "eval_runtime": 17.8732,
+ "eval_samples_per_second": 30.157,
+ "eval_steps_per_second": 15.106,
+ "step": 200
+ },
+ {
+ "entropy": 0.28792863376438615,
+ "epoch": 0.6936416184971098,
+ "grad_norm": 1.9609375,
+ "learning_rate": 7.271062271062271e-06,
+ "loss": 0.29235827922821045,
+ "mean_token_accuracy": 0.9050419509410859,
+ "num_tokens": 1959819.0,
+ "step": 210
+ },
+ {
+ "entropy": 0.28098181951791046,
+ "epoch": 0.726672171758877,
+ "grad_norm": 1.8515625,
+ "learning_rate": 7.087912087912089e-06,
+ "loss": 0.2915766954421997,
+ "mean_token_accuracy": 0.9068245775997639,
+ "num_tokens": 2059113.0,
+ "step": 220
+ },
+ {
+ "entropy": 0.26028320472687483,
+ "epoch": 0.7597027250206441,
+ "grad_norm": 2.0,
+ "learning_rate": 6.9047619047619055e-06,
+ "loss": 0.2654874324798584,
+ "mean_token_accuracy": 0.9116183832287789,
+ "num_tokens": 2145321.0,
+ "step": 230
+ },
+ {
+ "entropy": 0.2855058180168271,
+ "epoch": 0.7927332782824112,
+ "grad_norm": 1.84375,
+ "learning_rate": 6.721611721611723e-06,
+ "loss": 0.2998212337493896,
+ "mean_token_accuracy": 0.9046589516103267,
+ "num_tokens": 2240859.0,
+ "step": 240
+ },
+ {
+ "entropy": 0.2783821925520897,
+ "epoch": 0.8257638315441783,
+ "grad_norm": 1.9609375,
+ "learning_rate": 6.538461538461539e-06,
+ "loss": 0.28870720863342286,
+ "mean_token_accuracy": 0.9073394626379013,
+ "num_tokens": 2333798.0,
+ "step": 250
+ },
+ {
+ "entropy": 0.277759512513876,
+ "epoch": 0.8587943848059455,
+ "grad_norm": 1.7734375,
+ "learning_rate": 6.3553113553113555e-06,
+ "loss": 0.29060728549957277,
+ "mean_token_accuracy": 0.9055352889001369,
+ "num_tokens": 2429910.0,
+ "step": 260
+ },
+ {
+ "entropy": 0.2882726760581136,
+ "epoch": 0.8918249380677127,
+ "grad_norm": 1.703125,
+ "learning_rate": 6.172161172161173e-06,
+ "loss": 0.2988502740859985,
+ "mean_token_accuracy": 0.9038613699376583,
+ "num_tokens": 2524730.0,
+ "step": 270
+ },
+ {
+ "entropy": 0.2855747708119452,
+ "epoch": 0.9248554913294798,
+ "grad_norm": 1.90625,
+ "learning_rate": 5.989010989010989e-06,
+ "loss": 0.289182186126709,
+ "mean_token_accuracy": 0.9049803033471108,
+ "num_tokens": 2615802.0,
+ "step": 280
+ },
+ {
+ "entropy": 0.26923007648438213,
+ "epoch": 0.9578860445912469,
+ "grad_norm": 2.84375,
+ "learning_rate": 5.805860805860807e-06,
+ "loss": 0.27834036350250246,
+ "mean_token_accuracy": 0.9109352678060532,
+ "num_tokens": 2709073.0,
+ "step": 290
+ },
+ {
+ "entropy": 0.2844470454379916,
+ "epoch": 0.990916597853014,
+ "grad_norm": 1.640625,
+ "learning_rate": 5.6227106227106235e-06,
+ "loss": 0.28873960971832274,
+ "mean_token_accuracy": 0.9069667845964432,
+ "num_tokens": 2805135.0,
+ "step": 300
+ },
+ {
+ "epoch": 0.990916597853014,
+ "eval_entropy": 0.285272175570329,
+ "eval_loss": 0.28705036640167236,
+ "eval_mean_token_accuracy": 0.9062359571456909,
+ "eval_num_tokens": 2805135.0,
+ "eval_runtime": 17.9038,
+ "eval_samples_per_second": 30.105,
+ "eval_steps_per_second": 15.081,
+ "step": 300
+ },
+ {
+ "entropy": 0.2869852078266633,
+ "epoch": 1.023121387283237,
+ "grad_norm": 1.9375,
+ "learning_rate": 5.43956043956044e-06,
+ "loss": 0.2850792646408081,
+ "mean_token_accuracy": 0.9064911672702203,
+ "num_tokens": 2898984.0,
+ "step": 310
+ },
+ {
+ "entropy": 0.2757460119202733,
+ "epoch": 1.0561519405450042,
+ "grad_norm": 1.8359375,
+ "learning_rate": 5.256410256410257e-06,
+ "loss": 0.2779810905456543,
+ "mean_token_accuracy": 0.9096895463764667,
+ "num_tokens": 2991694.0,
+ "step": 320
+ },
+ {
+ "entropy": 0.2600112119689584,
+ "epoch": 1.0891824938067713,
+ "grad_norm": 2.0625,
+ "learning_rate": 5.0732600732600735e-06,
+ "loss": 0.2662019729614258,
+ "mean_token_accuracy": 0.912333931028843,
+ "num_tokens": 3078862.0,
+ "step": 330
+ },
+ {
+ "entropy": 0.27342566605657337,
+ "epoch": 1.1222130470685383,
+ "grad_norm": 1.921875,
+ "learning_rate": 4.890109890109891e-06,
+ "loss": 0.28096842765808105,
+ "mean_token_accuracy": 0.9091624312102795,
+ "num_tokens": 3174256.0,
+ "step": 340
+ },
+ {
+ "entropy": 0.2671937083825469,
+ "epoch": 1.1552436003303055,
+ "grad_norm": 1.75,
+ "learning_rate": 4.706959706959707e-06,
+ "loss": 0.2723649501800537,
+ "mean_token_accuracy": 0.9101433955132961,
+ "num_tokens": 3272477.0,
+ "step": 350
+ },
+ {
+ "entropy": 0.2617026660591364,
+ "epoch": 1.1882741535920727,
+ "grad_norm": 1.9453125,
+ "learning_rate": 4.523809523809524e-06,
+ "loss": 0.26422107219696045,
+ "mean_token_accuracy": 0.913913157582283,
+ "num_tokens": 3364137.0,
+ "step": 360
+ },
+ {
+ "entropy": 0.2595691759139299,
+ "epoch": 1.2213047068538398,
+ "grad_norm": 1.5390625,
+ "learning_rate": 4.340659340659341e-06,
+ "loss": 0.2714951276779175,
+ "mean_token_accuracy": 0.911636833101511,
+ "num_tokens": 3460217.0,
+ "step": 370
+ },
+ {
+ "entropy": 0.2933320404961705,
+ "epoch": 1.254335260115607,
+ "grad_norm": 1.9765625,
+ "learning_rate": 4.157509157509158e-06,
+ "loss": 0.29573276042938235,
+ "mean_token_accuracy": 0.902192335575819,
+ "num_tokens": 3557030.0,
+ "step": 380
+ },
+ {
+ "entropy": 0.2693369995802641,
+ "epoch": 1.287365813377374,
+ "grad_norm": 1.890625,
+ "learning_rate": 3.974358974358974e-06,
+ "loss": 0.27785425186157225,
+ "mean_token_accuracy": 0.9090337291359901,
+ "num_tokens": 3654159.0,
+ "step": 390
+ },
+ {
+ "entropy": 0.2908114368095994,
+ "epoch": 1.3203963666391412,
+ "grad_norm": 2.015625,
+ "learning_rate": 3.7912087912087915e-06,
+ "loss": 0.30535905361175536,
+ "mean_token_accuracy": 0.901054035872221,
+ "num_tokens": 3749664.0,
+ "step": 400
+ },
+ {
+ "epoch": 1.3203963666391412,
+ "eval_entropy": 0.2783401116176888,
+ "eval_loss": 0.28495287895202637,
+ "eval_mean_token_accuracy": 0.9069356251645971,
+ "eval_num_tokens": 3749664.0,
+ "eval_runtime": 17.8814,
+ "eval_samples_per_second": 30.143,
+ "eval_steps_per_second": 15.099,
+ "step": 400
+ },
+ {
+ "entropy": 0.2749379588291049,
+ "epoch": 1.3534269199009084,
+ "grad_norm": 1.84375,
+ "learning_rate": 3.6080586080586083e-06,
+ "loss": 0.27795455455780027,
+ "mean_token_accuracy": 0.9090839587152004,
+ "num_tokens": 3845841.0,
+ "step": 410
+ },
+ {
+ "entropy": 0.2859474029392004,
+ "epoch": 1.3864574731626755,
+ "grad_norm": 1.6796875,
+ "learning_rate": 3.424908424908425e-06,
+ "loss": 0.2886682987213135,
+ "mean_token_accuracy": 0.9059950634837151,
+ "num_tokens": 3942402.0,
+ "step": 420
+ },
+ {
+ "entropy": 0.2537162031978369,
+ "epoch": 1.4194880264244425,
+ "grad_norm": 1.9375,
+ "learning_rate": 3.2417582417582424e-06,
+ "loss": 0.25533475875854494,
+ "mean_token_accuracy": 0.9158934533596039,
+ "num_tokens": 4034122.0,
+ "step": 430
+ },
+ {
+ "entropy": 0.28255065642297267,
+ "epoch": 1.4525185796862097,
+ "grad_norm": 1.7734375,
+ "learning_rate": 3.058608058608059e-06,
+ "loss": 0.29817631244659426,
+ "mean_token_accuracy": 0.9040707617998123,
+ "num_tokens": 4127068.0,
+ "step": 440
+ },
+ {
+ "entropy": 0.2898306529968977,
+ "epoch": 1.4855491329479769,
+ "grad_norm": 1.703125,
+ "learning_rate": 2.8754578754578755e-06,
+ "loss": 0.2923931121826172,
+ "mean_token_accuracy": 0.9030195340514183,
+ "num_tokens": 4224435.0,
+ "step": 450
+ },
+ {
+ "entropy": 0.27209411058574917,
+ "epoch": 1.518579686209744,
+ "grad_norm": 1.7578125,
+ "learning_rate": 2.6923076923076923e-06,
+ "loss": 0.28646528720855713,
+ "mean_token_accuracy": 0.9095881894230843,
+ "num_tokens": 4321028.0,
+ "step": 460
+ },
+ {
+ "entropy": 0.280727200768888,
+ "epoch": 1.5516102394715112,
+ "grad_norm": 1.7421875,
+ "learning_rate": 2.509157509157509e-06,
+ "loss": 0.2891413688659668,
+ "mean_token_accuracy": 0.9067636162042618,
+ "num_tokens": 4414916.0,
+ "step": 470
+ },
+ {
+ "entropy": 0.27109829243272543,
+ "epoch": 1.5846407927332784,
+ "grad_norm": 1.8359375,
+ "learning_rate": 2.3260073260073264e-06,
+ "loss": 0.26625516414642336,
+ "mean_token_accuracy": 0.9099020563066006,
+ "num_tokens": 4505708.0,
+ "step": 480
+ },
+ {
+ "entropy": 0.2729395367205143,
+ "epoch": 1.6176713459950454,
+ "grad_norm": 1.7890625,
+ "learning_rate": 2.1428571428571427e-06,
+ "loss": 0.2779061317443848,
+ "mean_token_accuracy": 0.909747340530157,
+ "num_tokens": 4600244.0,
+ "step": 490
+ },
+ {
+ "entropy": 0.2682301852852106,
+ "epoch": 1.6507018992568125,
+ "grad_norm": 1.6953125,
+ "learning_rate": 1.95970695970696e-06,
+ "loss": 0.274446964263916,
+ "mean_token_accuracy": 0.9094639919698239,
+ "num_tokens": 4688147.0,
+ "step": 500
+ },
+ {
+ "epoch": 1.6507018992568125,
+ "eval_entropy": 0.2756172451708052,
+ "eval_loss": 0.2841346263885498,
+ "eval_mean_token_accuracy": 0.9072470570052111,
+ "eval_num_tokens": 4688147.0,
+ "eval_runtime": 17.9119,
+ "eval_samples_per_second": 30.092,
+ "eval_steps_per_second": 15.074,
+ "step": 500
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 606,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 2,
+ "save_steps": 100,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": false
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 5.028742352950886e+16,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-500/training_args.bin b/checkpoint-500/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..133cb6661b0124bb9db0d7d208d748fd736ea689
--- /dev/null
+++ b/checkpoint-500/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6696e1fef8bde5e22de2f5ae18c284f2b2d35a345157db5151571518385fb38b
+size 5713
diff --git a/checkpoint-600/chat_template.jinja b/checkpoint-600/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-600/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-600/config.json b/checkpoint-600/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-600/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-600/generation_config.json b/checkpoint-600/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-600/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-600/model.safetensors b/checkpoint-600/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..888c633f416471cd0dba703e4ed15cd967deaf00
--- /dev/null
+++ b/checkpoint-600/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:066f20885c5c9a598d45fbcd21aa7ab9081dffda3fb1930d573d7c3d99b5d998
+size 3441185608
diff --git a/checkpoint-600/optimizer.pt b/checkpoint-600/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..315889bc3c6c060ec3677e282757036eb6d14155
--- /dev/null
+++ b/checkpoint-600/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c75c57fe3c14c29f7a73acb2b44e60573aaf69cd744a4ce99770cdfdbe4de45e
+size 6882572207
diff --git a/checkpoint-600/rng_state.pth b/checkpoint-600/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..1fb544a66454d577cf2a3cb06c26493972f57952
--- /dev/null
+++ b/checkpoint-600/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fdc37bbd2e979f041dfbbb004a5c74bab6cdda159cb18116df728588515a9ef6
+size 14645
diff --git a/checkpoint-600/scheduler.pt b/checkpoint-600/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..e2645e63119eba5c7c9361da38b055b60e6d8985
--- /dev/null
+++ b/checkpoint-600/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:540d8ab45fe72a01988766f9263355239d34a4ff0ef13ddac7f1f0d1198ce9a9
+size 1465
diff --git a/checkpoint-600/tokenizer.json b/checkpoint-600/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-600/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-600/tokenizer_config.json b/checkpoint-600/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-600/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-600/trainer_state.json b/checkpoint-600/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..c4cf7cbe351f817a25fffdf26360f8e89d0ea1ce
--- /dev/null
+++ b/checkpoint-600/trainer_state.json
@@ -0,0 +1,700 @@
+{
+ "best_global_step": 600,
+ "best_metric": 0.28394654393196106,
+ "best_model_checkpoint": "/scratch/checkpoints/math_sft_v2/checkpoint-600",
+ "epoch": 1.981007431874484,
+ "eval_steps": 100,
+ "global_step": 600,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": 0.26208699569106103,
+ "epoch": 0.03303055326176713,
+ "grad_norm": 13.3125,
+ "learning_rate": 1.5e-06,
+ "loss": 0.8376299858093261,
+ "mean_token_accuracy": 0.8421687923371792,
+ "num_tokens": 98215.0,
+ "step": 10
+ },
+ {
+ "entropy": 0.2689803125336766,
+ "epoch": 0.06606110652353427,
+ "grad_norm": 11.5625,
+ "learning_rate": 3.1666666666666667e-06,
+ "loss": 0.853145980834961,
+ "mean_token_accuracy": 0.8399689704179764,
+ "num_tokens": 191821.0,
+ "step": 20
+ },
+ {
+ "entropy": 0.28777916859835384,
+ "epoch": 0.0990916597853014,
+ "grad_norm": 12.375,
+ "learning_rate": 4.833333333333333e-06,
+ "loss": 0.7679028034210205,
+ "mean_token_accuracy": 0.8431851923465729,
+ "num_tokens": 285491.0,
+ "step": 30
+ },
+ {
+ "entropy": 0.3033197607845068,
+ "epoch": 0.13212221304706853,
+ "grad_norm": 5.375,
+ "learning_rate": 6.5000000000000004e-06,
+ "loss": 0.6137509822845459,
+ "mean_token_accuracy": 0.8549904882907867,
+ "num_tokens": 381273.0,
+ "step": 40
+ },
+ {
+ "entropy": 0.34702248610556125,
+ "epoch": 0.16515276630883569,
+ "grad_norm": 4.25,
+ "learning_rate": 8.166666666666668e-06,
+ "loss": 0.5155063152313233,
+ "mean_token_accuracy": 0.8607264116406441,
+ "num_tokens": 476481.0,
+ "step": 50
+ },
+ {
+ "entropy": 0.35821260381489994,
+ "epoch": 0.1981833195706028,
+ "grad_norm": 2.890625,
+ "learning_rate": 9.833333333333333e-06,
+ "loss": 0.42441272735595703,
+ "mean_token_accuracy": 0.8774531938135623,
+ "num_tokens": 570979.0,
+ "step": 60
+ },
+ {
+ "entropy": 0.33337059412151576,
+ "epoch": 0.23121387283236994,
+ "grad_norm": 3.15625,
+ "learning_rate": 9.835164835164835e-06,
+ "loss": 0.34101109504699706,
+ "mean_token_accuracy": 0.8939957685768605,
+ "num_tokens": 664789.0,
+ "step": 70
+ },
+ {
+ "entropy": 0.30876336991786957,
+ "epoch": 0.26424442609413706,
+ "grad_norm": 1.8515625,
+ "learning_rate": 9.652014652014653e-06,
+ "loss": 0.32317628860473635,
+ "mean_token_accuracy": 0.8987478196620942,
+ "num_tokens": 764246.0,
+ "step": 80
+ },
+ {
+ "entropy": 0.30824985895305873,
+ "epoch": 0.2972749793559042,
+ "grad_norm": 2.15625,
+ "learning_rate": 9.46886446886447e-06,
+ "loss": 0.31807353496551516,
+ "mean_token_accuracy": 0.896314675360918,
+ "num_tokens": 855689.0,
+ "step": 90
+ },
+ {
+ "entropy": 0.31522620394825934,
+ "epoch": 0.33030553261767137,
+ "grad_norm": 1.8828125,
+ "learning_rate": 9.285714285714288e-06,
+ "loss": 0.338785982131958,
+ "mean_token_accuracy": 0.89400485008955,
+ "num_tokens": 949366.0,
+ "step": 100
+ },
+ {
+ "epoch": 0.33030553261767137,
+ "eval_entropy": 0.3022541454544774,
+ "eval_loss": 0.3146008253097534,
+ "eval_mean_token_accuracy": 0.8988911988558592,
+ "eval_num_tokens": 949366.0,
+ "eval_runtime": 17.914,
+ "eval_samples_per_second": 30.088,
+ "eval_steps_per_second": 15.072,
+ "step": 100
+ },
+ {
+ "entropy": 0.2985268572345376,
+ "epoch": 0.3633360858794385,
+ "grad_norm": 2.171875,
+ "learning_rate": 9.102564102564104e-06,
+ "loss": 0.31417596340179443,
+ "mean_token_accuracy": 0.89812126532197,
+ "num_tokens": 1040854.0,
+ "step": 110
+ },
+ {
+ "entropy": 0.3040375892072916,
+ "epoch": 0.3963666391412056,
+ "grad_norm": 1.9609375,
+ "learning_rate": 8.91941391941392e-06,
+ "loss": 0.32050628662109376,
+ "mean_token_accuracy": 0.8970061101019382,
+ "num_tokens": 1132424.0,
+ "step": 120
+ },
+ {
+ "entropy": 0.3125588171184063,
+ "epoch": 0.42939719240297275,
+ "grad_norm": 1.8203125,
+ "learning_rate": 8.736263736263737e-06,
+ "loss": 0.3176266193389893,
+ "mean_token_accuracy": 0.8971590586006641,
+ "num_tokens": 1221899.0,
+ "step": 130
+ },
+ {
+ "entropy": 0.312872576713562,
+ "epoch": 0.4624277456647399,
+ "grad_norm": 1.78125,
+ "learning_rate": 8.553113553113553e-06,
+ "loss": 0.3343791484832764,
+ "mean_token_accuracy": 0.8940604299306869,
+ "num_tokens": 1315052.0,
+ "step": 140
+ },
+ {
+ "entropy": 0.27830475065857174,
+ "epoch": 0.495458298926507,
+ "grad_norm": 2.0,
+ "learning_rate": 8.369963369963371e-06,
+ "loss": 0.28718521595001223,
+ "mean_token_accuracy": 0.9051955968141556,
+ "num_tokens": 1401947.0,
+ "step": 150
+ },
+ {
+ "entropy": 0.3005422970280051,
+ "epoch": 0.5284888521882741,
+ "grad_norm": 2.34375,
+ "learning_rate": 8.186813186813188e-06,
+ "loss": 0.31569790840148926,
+ "mean_token_accuracy": 0.8999212823808194,
+ "num_tokens": 1494107.0,
+ "step": 160
+ },
+ {
+ "entropy": 0.31200682315975425,
+ "epoch": 0.5615194054500413,
+ "grad_norm": 2.015625,
+ "learning_rate": 8.003663003663006e-06,
+ "loss": 0.32325453758239747,
+ "mean_token_accuracy": 0.8952137432992459,
+ "num_tokens": 1586474.0,
+ "step": 170
+ },
+ {
+ "entropy": 0.2929861258715391,
+ "epoch": 0.5945499587118084,
+ "grad_norm": 2.09375,
+ "learning_rate": 7.820512820512822e-06,
+ "loss": 0.29693677425384524,
+ "mean_token_accuracy": 0.9031261466443539,
+ "num_tokens": 1678162.0,
+ "step": 180
+ },
+ {
+ "entropy": 0.30670158881694076,
+ "epoch": 0.6275805119735756,
+ "grad_norm": 1.828125,
+ "learning_rate": 7.637362637362638e-06,
+ "loss": 0.31567308902740476,
+ "mean_token_accuracy": 0.8983482263982296,
+ "num_tokens": 1781401.0,
+ "step": 190
+ },
+ {
+ "entropy": 0.2866023685783148,
+ "epoch": 0.6606110652353427,
+ "grad_norm": 1.6953125,
+ "learning_rate": 7.454212454212455e-06,
+ "loss": 0.29333360195159913,
+ "mean_token_accuracy": 0.904055442661047,
+ "num_tokens": 1868038.0,
+ "step": 200
+ },
+ {
+ "epoch": 0.6606110652353427,
+ "eval_entropy": 0.2921138784951634,
+ "eval_loss": 0.29292193055152893,
+ "eval_mean_token_accuracy": 0.904505420172656,
+ "eval_num_tokens": 1868038.0,
+ "eval_runtime": 17.8732,
+ "eval_samples_per_second": 30.157,
+ "eval_steps_per_second": 15.106,
+ "step": 200
+ },
+ {
+ "entropy": 0.28792863376438615,
+ "epoch": 0.6936416184971098,
+ "grad_norm": 1.9609375,
+ "learning_rate": 7.271062271062271e-06,
+ "loss": 0.29235827922821045,
+ "mean_token_accuracy": 0.9050419509410859,
+ "num_tokens": 1959819.0,
+ "step": 210
+ },
+ {
+ "entropy": 0.28098181951791046,
+ "epoch": 0.726672171758877,
+ "grad_norm": 1.8515625,
+ "learning_rate": 7.087912087912089e-06,
+ "loss": 0.2915766954421997,
+ "mean_token_accuracy": 0.9068245775997639,
+ "num_tokens": 2059113.0,
+ "step": 220
+ },
+ {
+ "entropy": 0.26028320472687483,
+ "epoch": 0.7597027250206441,
+ "grad_norm": 2.0,
+ "learning_rate": 6.9047619047619055e-06,
+ "loss": 0.2654874324798584,
+ "mean_token_accuracy": 0.9116183832287789,
+ "num_tokens": 2145321.0,
+ "step": 230
+ },
+ {
+ "entropy": 0.2855058180168271,
+ "epoch": 0.7927332782824112,
+ "grad_norm": 1.84375,
+ "learning_rate": 6.721611721611723e-06,
+ "loss": 0.2998212337493896,
+ "mean_token_accuracy": 0.9046589516103267,
+ "num_tokens": 2240859.0,
+ "step": 240
+ },
+ {
+ "entropy": 0.2783821925520897,
+ "epoch": 0.8257638315441783,
+ "grad_norm": 1.9609375,
+ "learning_rate": 6.538461538461539e-06,
+ "loss": 0.28870720863342286,
+ "mean_token_accuracy": 0.9073394626379013,
+ "num_tokens": 2333798.0,
+ "step": 250
+ },
+ {
+ "entropy": 0.277759512513876,
+ "epoch": 0.8587943848059455,
+ "grad_norm": 1.7734375,
+ "learning_rate": 6.3553113553113555e-06,
+ "loss": 0.29060728549957277,
+ "mean_token_accuracy": 0.9055352889001369,
+ "num_tokens": 2429910.0,
+ "step": 260
+ },
+ {
+ "entropy": 0.2882726760581136,
+ "epoch": 0.8918249380677127,
+ "grad_norm": 1.703125,
+ "learning_rate": 6.172161172161173e-06,
+ "loss": 0.2988502740859985,
+ "mean_token_accuracy": 0.9038613699376583,
+ "num_tokens": 2524730.0,
+ "step": 270
+ },
+ {
+ "entropy": 0.2855747708119452,
+ "epoch": 0.9248554913294798,
+ "grad_norm": 1.90625,
+ "learning_rate": 5.989010989010989e-06,
+ "loss": 0.289182186126709,
+ "mean_token_accuracy": 0.9049803033471108,
+ "num_tokens": 2615802.0,
+ "step": 280
+ },
+ {
+ "entropy": 0.26923007648438213,
+ "epoch": 0.9578860445912469,
+ "grad_norm": 2.84375,
+ "learning_rate": 5.805860805860807e-06,
+ "loss": 0.27834036350250246,
+ "mean_token_accuracy": 0.9109352678060532,
+ "num_tokens": 2709073.0,
+ "step": 290
+ },
+ {
+ "entropy": 0.2844470454379916,
+ "epoch": 0.990916597853014,
+ "grad_norm": 1.640625,
+ "learning_rate": 5.6227106227106235e-06,
+ "loss": 0.28873960971832274,
+ "mean_token_accuracy": 0.9069667845964432,
+ "num_tokens": 2805135.0,
+ "step": 300
+ },
+ {
+ "epoch": 0.990916597853014,
+ "eval_entropy": 0.285272175570329,
+ "eval_loss": 0.28705036640167236,
+ "eval_mean_token_accuracy": 0.9062359571456909,
+ "eval_num_tokens": 2805135.0,
+ "eval_runtime": 17.9038,
+ "eval_samples_per_second": 30.105,
+ "eval_steps_per_second": 15.081,
+ "step": 300
+ },
+ {
+ "entropy": 0.2869852078266633,
+ "epoch": 1.023121387283237,
+ "grad_norm": 1.9375,
+ "learning_rate": 5.43956043956044e-06,
+ "loss": 0.2850792646408081,
+ "mean_token_accuracy": 0.9064911672702203,
+ "num_tokens": 2898984.0,
+ "step": 310
+ },
+ {
+ "entropy": 0.2757460119202733,
+ "epoch": 1.0561519405450042,
+ "grad_norm": 1.8359375,
+ "learning_rate": 5.256410256410257e-06,
+ "loss": 0.2779810905456543,
+ "mean_token_accuracy": 0.9096895463764667,
+ "num_tokens": 2991694.0,
+ "step": 320
+ },
+ {
+ "entropy": 0.2600112119689584,
+ "epoch": 1.0891824938067713,
+ "grad_norm": 2.0625,
+ "learning_rate": 5.0732600732600735e-06,
+ "loss": 0.2662019729614258,
+ "mean_token_accuracy": 0.912333931028843,
+ "num_tokens": 3078862.0,
+ "step": 330
+ },
+ {
+ "entropy": 0.27342566605657337,
+ "epoch": 1.1222130470685383,
+ "grad_norm": 1.921875,
+ "learning_rate": 4.890109890109891e-06,
+ "loss": 0.28096842765808105,
+ "mean_token_accuracy": 0.9091624312102795,
+ "num_tokens": 3174256.0,
+ "step": 340
+ },
+ {
+ "entropy": 0.2671937083825469,
+ "epoch": 1.1552436003303055,
+ "grad_norm": 1.75,
+ "learning_rate": 4.706959706959707e-06,
+ "loss": 0.2723649501800537,
+ "mean_token_accuracy": 0.9101433955132961,
+ "num_tokens": 3272477.0,
+ "step": 350
+ },
+ {
+ "entropy": 0.2617026660591364,
+ "epoch": 1.1882741535920727,
+ "grad_norm": 1.9453125,
+ "learning_rate": 4.523809523809524e-06,
+ "loss": 0.26422107219696045,
+ "mean_token_accuracy": 0.913913157582283,
+ "num_tokens": 3364137.0,
+ "step": 360
+ },
+ {
+ "entropy": 0.2595691759139299,
+ "epoch": 1.2213047068538398,
+ "grad_norm": 1.5390625,
+ "learning_rate": 4.340659340659341e-06,
+ "loss": 0.2714951276779175,
+ "mean_token_accuracy": 0.911636833101511,
+ "num_tokens": 3460217.0,
+ "step": 370
+ },
+ {
+ "entropy": 0.2933320404961705,
+ "epoch": 1.254335260115607,
+ "grad_norm": 1.9765625,
+ "learning_rate": 4.157509157509158e-06,
+ "loss": 0.29573276042938235,
+ "mean_token_accuracy": 0.902192335575819,
+ "num_tokens": 3557030.0,
+ "step": 380
+ },
+ {
+ "entropy": 0.2693369995802641,
+ "epoch": 1.287365813377374,
+ "grad_norm": 1.890625,
+ "learning_rate": 3.974358974358974e-06,
+ "loss": 0.27785425186157225,
+ "mean_token_accuracy": 0.9090337291359901,
+ "num_tokens": 3654159.0,
+ "step": 390
+ },
+ {
+ "entropy": 0.2908114368095994,
+ "epoch": 1.3203963666391412,
+ "grad_norm": 2.015625,
+ "learning_rate": 3.7912087912087915e-06,
+ "loss": 0.30535905361175536,
+ "mean_token_accuracy": 0.901054035872221,
+ "num_tokens": 3749664.0,
+ "step": 400
+ },
+ {
+ "epoch": 1.3203963666391412,
+ "eval_entropy": 0.2783401116176888,
+ "eval_loss": 0.28495287895202637,
+ "eval_mean_token_accuracy": 0.9069356251645971,
+ "eval_num_tokens": 3749664.0,
+ "eval_runtime": 17.8814,
+ "eval_samples_per_second": 30.143,
+ "eval_steps_per_second": 15.099,
+ "step": 400
+ },
+ {
+ "entropy": 0.2749379588291049,
+ "epoch": 1.3534269199009084,
+ "grad_norm": 1.84375,
+ "learning_rate": 3.6080586080586083e-06,
+ "loss": 0.27795455455780027,
+ "mean_token_accuracy": 0.9090839587152004,
+ "num_tokens": 3845841.0,
+ "step": 410
+ },
+ {
+ "entropy": 0.2859474029392004,
+ "epoch": 1.3864574731626755,
+ "grad_norm": 1.6796875,
+ "learning_rate": 3.424908424908425e-06,
+ "loss": 0.2886682987213135,
+ "mean_token_accuracy": 0.9059950634837151,
+ "num_tokens": 3942402.0,
+ "step": 420
+ },
+ {
+ "entropy": 0.2537162031978369,
+ "epoch": 1.4194880264244425,
+ "grad_norm": 1.9375,
+ "learning_rate": 3.2417582417582424e-06,
+ "loss": 0.25533475875854494,
+ "mean_token_accuracy": 0.9158934533596039,
+ "num_tokens": 4034122.0,
+ "step": 430
+ },
+ {
+ "entropy": 0.28255065642297267,
+ "epoch": 1.4525185796862097,
+ "grad_norm": 1.7734375,
+ "learning_rate": 3.058608058608059e-06,
+ "loss": 0.29817631244659426,
+ "mean_token_accuracy": 0.9040707617998123,
+ "num_tokens": 4127068.0,
+ "step": 440
+ },
+ {
+ "entropy": 0.2898306529968977,
+ "epoch": 1.4855491329479769,
+ "grad_norm": 1.703125,
+ "learning_rate": 2.8754578754578755e-06,
+ "loss": 0.2923931121826172,
+ "mean_token_accuracy": 0.9030195340514183,
+ "num_tokens": 4224435.0,
+ "step": 450
+ },
+ {
+ "entropy": 0.27209411058574917,
+ "epoch": 1.518579686209744,
+ "grad_norm": 1.7578125,
+ "learning_rate": 2.6923076923076923e-06,
+ "loss": 0.28646528720855713,
+ "mean_token_accuracy": 0.9095881894230843,
+ "num_tokens": 4321028.0,
+ "step": 460
+ },
+ {
+ "entropy": 0.280727200768888,
+ "epoch": 1.5516102394715112,
+ "grad_norm": 1.7421875,
+ "learning_rate": 2.509157509157509e-06,
+ "loss": 0.2891413688659668,
+ "mean_token_accuracy": 0.9067636162042618,
+ "num_tokens": 4414916.0,
+ "step": 470
+ },
+ {
+ "entropy": 0.27109829243272543,
+ "epoch": 1.5846407927332784,
+ "grad_norm": 1.8359375,
+ "learning_rate": 2.3260073260073264e-06,
+ "loss": 0.26625516414642336,
+ "mean_token_accuracy": 0.9099020563066006,
+ "num_tokens": 4505708.0,
+ "step": 480
+ },
+ {
+ "entropy": 0.2729395367205143,
+ "epoch": 1.6176713459950454,
+ "grad_norm": 1.7890625,
+ "learning_rate": 2.1428571428571427e-06,
+ "loss": 0.2779061317443848,
+ "mean_token_accuracy": 0.909747340530157,
+ "num_tokens": 4600244.0,
+ "step": 490
+ },
+ {
+ "entropy": 0.2682301852852106,
+ "epoch": 1.6507018992568125,
+ "grad_norm": 1.6953125,
+ "learning_rate": 1.95970695970696e-06,
+ "loss": 0.274446964263916,
+ "mean_token_accuracy": 0.9094639919698239,
+ "num_tokens": 4688147.0,
+ "step": 500
+ },
+ {
+ "epoch": 1.6507018992568125,
+ "eval_entropy": 0.2756172451708052,
+ "eval_loss": 0.2841346263885498,
+ "eval_mean_token_accuracy": 0.9072470570052111,
+ "eval_num_tokens": 4688147.0,
+ "eval_runtime": 17.9119,
+ "eval_samples_per_second": 30.092,
+ "eval_steps_per_second": 15.074,
+ "step": 500
+ },
+ {
+ "entropy": 0.26342453975230457,
+ "epoch": 1.6837324525185797,
+ "grad_norm": 1.890625,
+ "learning_rate": 1.7765567765567768e-06,
+ "loss": 0.26432271003723146,
+ "mean_token_accuracy": 0.9114409819245338,
+ "num_tokens": 4780124.0,
+ "step": 510
+ },
+ {
+ "entropy": 0.28657868495211003,
+ "epoch": 1.7167630057803467,
+ "grad_norm": 1.7734375,
+ "learning_rate": 1.5934065934065933e-06,
+ "loss": 0.2983593225479126,
+ "mean_token_accuracy": 0.9028328903019428,
+ "num_tokens": 4874633.0,
+ "step": 520
+ },
+ {
+ "entropy": 0.25633321776986123,
+ "epoch": 1.7497935590421139,
+ "grad_norm": 1.8359375,
+ "learning_rate": 1.4102564102564104e-06,
+ "loss": 0.2606488227844238,
+ "mean_token_accuracy": 0.9153999648988247,
+ "num_tokens": 4967263.0,
+ "step": 530
+ },
+ {
+ "entropy": 0.26120030507445335,
+ "epoch": 1.782824112303881,
+ "grad_norm": 2.078125,
+ "learning_rate": 1.2271062271062271e-06,
+ "loss": 0.2706313610076904,
+ "mean_token_accuracy": 0.9130682416260243,
+ "num_tokens": 5056736.0,
+ "step": 540
+ },
+ {
+ "entropy": 0.27009565513581035,
+ "epoch": 1.8158546655656482,
+ "grad_norm": 1.8828125,
+ "learning_rate": 1.0439560439560442e-06,
+ "loss": 0.277858567237854,
+ "mean_token_accuracy": 0.9087999157607556,
+ "num_tokens": 5150617.0,
+ "step": 550
+ },
+ {
+ "entropy": 0.2819518899545074,
+ "epoch": 1.8488852188274154,
+ "grad_norm": 1.9609375,
+ "learning_rate": 8.608058608058609e-07,
+ "loss": 0.28743035793304444,
+ "mean_token_accuracy": 0.9067244723439216,
+ "num_tokens": 5243538.0,
+ "step": 560
+ },
+ {
+ "entropy": 0.2858045116066933,
+ "epoch": 1.8819157720891826,
+ "grad_norm": 2.203125,
+ "learning_rate": 6.776556776556777e-07,
+ "loss": 0.2939825296401978,
+ "mean_token_accuracy": 0.904579871147871,
+ "num_tokens": 5331441.0,
+ "step": 570
+ },
+ {
+ "entropy": 0.2926233192905784,
+ "epoch": 1.9149463253509498,
+ "grad_norm": 2.3125,
+ "learning_rate": 4.945054945054946e-07,
+ "loss": 0.29961700439453126,
+ "mean_token_accuracy": 0.9033655568957328,
+ "num_tokens": 5426734.0,
+ "step": 580
+ },
+ {
+ "entropy": 0.2706678254529834,
+ "epoch": 1.9479768786127167,
+ "grad_norm": 2.03125,
+ "learning_rate": 3.113553113553114e-07,
+ "loss": 0.27953526973724363,
+ "mean_token_accuracy": 0.9093762949109078,
+ "num_tokens": 5518190.0,
+ "step": 590
+ },
+ {
+ "entropy": 0.27056904807686805,
+ "epoch": 1.981007431874484,
+ "grad_norm": 2.09375,
+ "learning_rate": 1.282051282051282e-07,
+ "loss": 0.2764536619186401,
+ "mean_token_accuracy": 0.9110228553414345,
+ "num_tokens": 5609238.0,
+ "step": 600
+ },
+ {
+ "epoch": 1.981007431874484,
+ "eval_entropy": 0.2757842331572815,
+ "eval_loss": 0.28394654393196106,
+ "eval_mean_token_accuracy": 0.9073928307603907,
+ "eval_num_tokens": 5609238.0,
+ "eval_runtime": 18.1138,
+ "eval_samples_per_second": 29.756,
+ "eval_steps_per_second": 14.906,
+ "step": 600
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 606,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 2,
+ "save_steps": 100,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": false
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 6.018332557481165e+16,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-600/training_args.bin b/checkpoint-600/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..133cb6661b0124bb9db0d7d208d748fd736ea689
--- /dev/null
+++ b/checkpoint-600/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6696e1fef8bde5e22de2f5ae18c284f2b2d35a345157db5151571518385fb38b
+size 5713
diff --git a/checkpoint-606/chat_template.jinja b/checkpoint-606/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-606/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-606/config.json b/checkpoint-606/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-606/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-606/generation_config.json b/checkpoint-606/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-606/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-606/model.safetensors b/checkpoint-606/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..b08d1501dfdc93522108efc90dd1cd038d65cc7a
--- /dev/null
+++ b/checkpoint-606/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b2e0bbbd38e2572fcb7a7cdd2bf58165f2cb2a115544c68b132848cd7973d106
+size 3441185608
diff --git a/checkpoint-606/optimizer.pt b/checkpoint-606/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..590b9fb7b131560c1b58842b49379b273c5a6449
--- /dev/null
+++ b/checkpoint-606/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc3d8578da8d88239c2024e2dd1da1012e2173f003e746efcc15b17915d0a03c
+size 6882572207
diff --git a/checkpoint-606/rng_state.pth b/checkpoint-606/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..eed58b88838bddfd46b410984c9cbddf403b1437
--- /dev/null
+++ b/checkpoint-606/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:363c5df1543d2c82b2f13164f35bdd0367ceb32e7fa1b2f67c19df073a08b17b
+size 14645
diff --git a/checkpoint-606/scheduler.pt b/checkpoint-606/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..f182dc5e2546b5fdccbdc22b90eb16a565c67e6c
--- /dev/null
+++ b/checkpoint-606/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ef108e31778ee002976f58f4a1ef7b3f1228ac6f0b9152b3822e08d82d99376b
+size 1465
diff --git a/checkpoint-606/tokenizer.json b/checkpoint-606/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-606/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-606/tokenizer_config.json b/checkpoint-606/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-606/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-606/trainer_state.json b/checkpoint-606/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..a4ec67787e726511e6fbf586ff5bd33e585236d7
--- /dev/null
+++ b/checkpoint-606/trainer_state.json
@@ -0,0 +1,711 @@
+{
+ "best_global_step": 600,
+ "best_metric": 0.28394654393196106,
+ "best_model_checkpoint": "/scratch/checkpoints/math_sft_v2/checkpoint-600",
+ "epoch": 2.0,
+ "eval_steps": 100,
+ "global_step": 606,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": 0.26208699569106103,
+ "epoch": 0.03303055326176713,
+ "grad_norm": 13.3125,
+ "learning_rate": 1.5e-06,
+ "loss": 0.8376299858093261,
+ "mean_token_accuracy": 0.8421687923371792,
+ "num_tokens": 98215.0,
+ "step": 10
+ },
+ {
+ "entropy": 0.2689803125336766,
+ "epoch": 0.06606110652353427,
+ "grad_norm": 11.5625,
+ "learning_rate": 3.1666666666666667e-06,
+ "loss": 0.853145980834961,
+ "mean_token_accuracy": 0.8399689704179764,
+ "num_tokens": 191821.0,
+ "step": 20
+ },
+ {
+ "entropy": 0.28777916859835384,
+ "epoch": 0.0990916597853014,
+ "grad_norm": 12.375,
+ "learning_rate": 4.833333333333333e-06,
+ "loss": 0.7679028034210205,
+ "mean_token_accuracy": 0.8431851923465729,
+ "num_tokens": 285491.0,
+ "step": 30
+ },
+ {
+ "entropy": 0.3033197607845068,
+ "epoch": 0.13212221304706853,
+ "grad_norm": 5.375,
+ "learning_rate": 6.5000000000000004e-06,
+ "loss": 0.6137509822845459,
+ "mean_token_accuracy": 0.8549904882907867,
+ "num_tokens": 381273.0,
+ "step": 40
+ },
+ {
+ "entropy": 0.34702248610556125,
+ "epoch": 0.16515276630883569,
+ "grad_norm": 4.25,
+ "learning_rate": 8.166666666666668e-06,
+ "loss": 0.5155063152313233,
+ "mean_token_accuracy": 0.8607264116406441,
+ "num_tokens": 476481.0,
+ "step": 50
+ },
+ {
+ "entropy": 0.35821260381489994,
+ "epoch": 0.1981833195706028,
+ "grad_norm": 2.890625,
+ "learning_rate": 9.833333333333333e-06,
+ "loss": 0.42441272735595703,
+ "mean_token_accuracy": 0.8774531938135623,
+ "num_tokens": 570979.0,
+ "step": 60
+ },
+ {
+ "entropy": 0.33337059412151576,
+ "epoch": 0.23121387283236994,
+ "grad_norm": 3.15625,
+ "learning_rate": 9.835164835164835e-06,
+ "loss": 0.34101109504699706,
+ "mean_token_accuracy": 0.8939957685768605,
+ "num_tokens": 664789.0,
+ "step": 70
+ },
+ {
+ "entropy": 0.30876336991786957,
+ "epoch": 0.26424442609413706,
+ "grad_norm": 1.8515625,
+ "learning_rate": 9.652014652014653e-06,
+ "loss": 0.32317628860473635,
+ "mean_token_accuracy": 0.8987478196620942,
+ "num_tokens": 764246.0,
+ "step": 80
+ },
+ {
+ "entropy": 0.30824985895305873,
+ "epoch": 0.2972749793559042,
+ "grad_norm": 2.15625,
+ "learning_rate": 9.46886446886447e-06,
+ "loss": 0.31807353496551516,
+ "mean_token_accuracy": 0.896314675360918,
+ "num_tokens": 855689.0,
+ "step": 90
+ },
+ {
+ "entropy": 0.31522620394825934,
+ "epoch": 0.33030553261767137,
+ "grad_norm": 1.8828125,
+ "learning_rate": 9.285714285714288e-06,
+ "loss": 0.338785982131958,
+ "mean_token_accuracy": 0.89400485008955,
+ "num_tokens": 949366.0,
+ "step": 100
+ },
+ {
+ "epoch": 0.33030553261767137,
+ "eval_entropy": 0.3022541454544774,
+ "eval_loss": 0.3146008253097534,
+ "eval_mean_token_accuracy": 0.8988911988558592,
+ "eval_num_tokens": 949366.0,
+ "eval_runtime": 17.914,
+ "eval_samples_per_second": 30.088,
+ "eval_steps_per_second": 15.072,
+ "step": 100
+ },
+ {
+ "entropy": 0.2985268572345376,
+ "epoch": 0.3633360858794385,
+ "grad_norm": 2.171875,
+ "learning_rate": 9.102564102564104e-06,
+ "loss": 0.31417596340179443,
+ "mean_token_accuracy": 0.89812126532197,
+ "num_tokens": 1040854.0,
+ "step": 110
+ },
+ {
+ "entropy": 0.3040375892072916,
+ "epoch": 0.3963666391412056,
+ "grad_norm": 1.9609375,
+ "learning_rate": 8.91941391941392e-06,
+ "loss": 0.32050628662109376,
+ "mean_token_accuracy": 0.8970061101019382,
+ "num_tokens": 1132424.0,
+ "step": 120
+ },
+ {
+ "entropy": 0.3125588171184063,
+ "epoch": 0.42939719240297275,
+ "grad_norm": 1.8203125,
+ "learning_rate": 8.736263736263737e-06,
+ "loss": 0.3176266193389893,
+ "mean_token_accuracy": 0.8971590586006641,
+ "num_tokens": 1221899.0,
+ "step": 130
+ },
+ {
+ "entropy": 0.312872576713562,
+ "epoch": 0.4624277456647399,
+ "grad_norm": 1.78125,
+ "learning_rate": 8.553113553113553e-06,
+ "loss": 0.3343791484832764,
+ "mean_token_accuracy": 0.8940604299306869,
+ "num_tokens": 1315052.0,
+ "step": 140
+ },
+ {
+ "entropy": 0.27830475065857174,
+ "epoch": 0.495458298926507,
+ "grad_norm": 2.0,
+ "learning_rate": 8.369963369963371e-06,
+ "loss": 0.28718521595001223,
+ "mean_token_accuracy": 0.9051955968141556,
+ "num_tokens": 1401947.0,
+ "step": 150
+ },
+ {
+ "entropy": 0.3005422970280051,
+ "epoch": 0.5284888521882741,
+ "grad_norm": 2.34375,
+ "learning_rate": 8.186813186813188e-06,
+ "loss": 0.31569790840148926,
+ "mean_token_accuracy": 0.8999212823808194,
+ "num_tokens": 1494107.0,
+ "step": 160
+ },
+ {
+ "entropy": 0.31200682315975425,
+ "epoch": 0.5615194054500413,
+ "grad_norm": 2.015625,
+ "learning_rate": 8.003663003663006e-06,
+ "loss": 0.32325453758239747,
+ "mean_token_accuracy": 0.8952137432992459,
+ "num_tokens": 1586474.0,
+ "step": 170
+ },
+ {
+ "entropy": 0.2929861258715391,
+ "epoch": 0.5945499587118084,
+ "grad_norm": 2.09375,
+ "learning_rate": 7.820512820512822e-06,
+ "loss": 0.29693677425384524,
+ "mean_token_accuracy": 0.9031261466443539,
+ "num_tokens": 1678162.0,
+ "step": 180
+ },
+ {
+ "entropy": 0.30670158881694076,
+ "epoch": 0.6275805119735756,
+ "grad_norm": 1.828125,
+ "learning_rate": 7.637362637362638e-06,
+ "loss": 0.31567308902740476,
+ "mean_token_accuracy": 0.8983482263982296,
+ "num_tokens": 1781401.0,
+ "step": 190
+ },
+ {
+ "entropy": 0.2866023685783148,
+ "epoch": 0.6606110652353427,
+ "grad_norm": 1.6953125,
+ "learning_rate": 7.454212454212455e-06,
+ "loss": 0.29333360195159913,
+ "mean_token_accuracy": 0.904055442661047,
+ "num_tokens": 1868038.0,
+ "step": 200
+ },
+ {
+ "epoch": 0.6606110652353427,
+ "eval_entropy": 0.2921138784951634,
+ "eval_loss": 0.29292193055152893,
+ "eval_mean_token_accuracy": 0.904505420172656,
+ "eval_num_tokens": 1868038.0,
+ "eval_runtime": 17.8732,
+ "eval_samples_per_second": 30.157,
+ "eval_steps_per_second": 15.106,
+ "step": 200
+ },
+ {
+ "entropy": 0.28792863376438615,
+ "epoch": 0.6936416184971098,
+ "grad_norm": 1.9609375,
+ "learning_rate": 7.271062271062271e-06,
+ "loss": 0.29235827922821045,
+ "mean_token_accuracy": 0.9050419509410859,
+ "num_tokens": 1959819.0,
+ "step": 210
+ },
+ {
+ "entropy": 0.28098181951791046,
+ "epoch": 0.726672171758877,
+ "grad_norm": 1.8515625,
+ "learning_rate": 7.087912087912089e-06,
+ "loss": 0.2915766954421997,
+ "mean_token_accuracy": 0.9068245775997639,
+ "num_tokens": 2059113.0,
+ "step": 220
+ },
+ {
+ "entropy": 0.26028320472687483,
+ "epoch": 0.7597027250206441,
+ "grad_norm": 2.0,
+ "learning_rate": 6.9047619047619055e-06,
+ "loss": 0.2654874324798584,
+ "mean_token_accuracy": 0.9116183832287789,
+ "num_tokens": 2145321.0,
+ "step": 230
+ },
+ {
+ "entropy": 0.2855058180168271,
+ "epoch": 0.7927332782824112,
+ "grad_norm": 1.84375,
+ "learning_rate": 6.721611721611723e-06,
+ "loss": 0.2998212337493896,
+ "mean_token_accuracy": 0.9046589516103267,
+ "num_tokens": 2240859.0,
+ "step": 240
+ },
+ {
+ "entropy": 0.2783821925520897,
+ "epoch": 0.8257638315441783,
+ "grad_norm": 1.9609375,
+ "learning_rate": 6.538461538461539e-06,
+ "loss": 0.28870720863342286,
+ "mean_token_accuracy": 0.9073394626379013,
+ "num_tokens": 2333798.0,
+ "step": 250
+ },
+ {
+ "entropy": 0.277759512513876,
+ "epoch": 0.8587943848059455,
+ "grad_norm": 1.7734375,
+ "learning_rate": 6.3553113553113555e-06,
+ "loss": 0.29060728549957277,
+ "mean_token_accuracy": 0.9055352889001369,
+ "num_tokens": 2429910.0,
+ "step": 260
+ },
+ {
+ "entropy": 0.2882726760581136,
+ "epoch": 0.8918249380677127,
+ "grad_norm": 1.703125,
+ "learning_rate": 6.172161172161173e-06,
+ "loss": 0.2988502740859985,
+ "mean_token_accuracy": 0.9038613699376583,
+ "num_tokens": 2524730.0,
+ "step": 270
+ },
+ {
+ "entropy": 0.2855747708119452,
+ "epoch": 0.9248554913294798,
+ "grad_norm": 1.90625,
+ "learning_rate": 5.989010989010989e-06,
+ "loss": 0.289182186126709,
+ "mean_token_accuracy": 0.9049803033471108,
+ "num_tokens": 2615802.0,
+ "step": 280
+ },
+ {
+ "entropy": 0.26923007648438213,
+ "epoch": 0.9578860445912469,
+ "grad_norm": 2.84375,
+ "learning_rate": 5.805860805860807e-06,
+ "loss": 0.27834036350250246,
+ "mean_token_accuracy": 0.9109352678060532,
+ "num_tokens": 2709073.0,
+ "step": 290
+ },
+ {
+ "entropy": 0.2844470454379916,
+ "epoch": 0.990916597853014,
+ "grad_norm": 1.640625,
+ "learning_rate": 5.6227106227106235e-06,
+ "loss": 0.28873960971832274,
+ "mean_token_accuracy": 0.9069667845964432,
+ "num_tokens": 2805135.0,
+ "step": 300
+ },
+ {
+ "epoch": 0.990916597853014,
+ "eval_entropy": 0.285272175570329,
+ "eval_loss": 0.28705036640167236,
+ "eval_mean_token_accuracy": 0.9062359571456909,
+ "eval_num_tokens": 2805135.0,
+ "eval_runtime": 17.9038,
+ "eval_samples_per_second": 30.105,
+ "eval_steps_per_second": 15.081,
+ "step": 300
+ },
+ {
+ "entropy": 0.2869852078266633,
+ "epoch": 1.023121387283237,
+ "grad_norm": 1.9375,
+ "learning_rate": 5.43956043956044e-06,
+ "loss": 0.2850792646408081,
+ "mean_token_accuracy": 0.9064911672702203,
+ "num_tokens": 2898984.0,
+ "step": 310
+ },
+ {
+ "entropy": 0.2757460119202733,
+ "epoch": 1.0561519405450042,
+ "grad_norm": 1.8359375,
+ "learning_rate": 5.256410256410257e-06,
+ "loss": 0.2779810905456543,
+ "mean_token_accuracy": 0.9096895463764667,
+ "num_tokens": 2991694.0,
+ "step": 320
+ },
+ {
+ "entropy": 0.2600112119689584,
+ "epoch": 1.0891824938067713,
+ "grad_norm": 2.0625,
+ "learning_rate": 5.0732600732600735e-06,
+ "loss": 0.2662019729614258,
+ "mean_token_accuracy": 0.912333931028843,
+ "num_tokens": 3078862.0,
+ "step": 330
+ },
+ {
+ "entropy": 0.27342566605657337,
+ "epoch": 1.1222130470685383,
+ "grad_norm": 1.921875,
+ "learning_rate": 4.890109890109891e-06,
+ "loss": 0.28096842765808105,
+ "mean_token_accuracy": 0.9091624312102795,
+ "num_tokens": 3174256.0,
+ "step": 340
+ },
+ {
+ "entropy": 0.2671937083825469,
+ "epoch": 1.1552436003303055,
+ "grad_norm": 1.75,
+ "learning_rate": 4.706959706959707e-06,
+ "loss": 0.2723649501800537,
+ "mean_token_accuracy": 0.9101433955132961,
+ "num_tokens": 3272477.0,
+ "step": 350
+ },
+ {
+ "entropy": 0.2617026660591364,
+ "epoch": 1.1882741535920727,
+ "grad_norm": 1.9453125,
+ "learning_rate": 4.523809523809524e-06,
+ "loss": 0.26422107219696045,
+ "mean_token_accuracy": 0.913913157582283,
+ "num_tokens": 3364137.0,
+ "step": 360
+ },
+ {
+ "entropy": 0.2595691759139299,
+ "epoch": 1.2213047068538398,
+ "grad_norm": 1.5390625,
+ "learning_rate": 4.340659340659341e-06,
+ "loss": 0.2714951276779175,
+ "mean_token_accuracy": 0.911636833101511,
+ "num_tokens": 3460217.0,
+ "step": 370
+ },
+ {
+ "entropy": 0.2933320404961705,
+ "epoch": 1.254335260115607,
+ "grad_norm": 1.9765625,
+ "learning_rate": 4.157509157509158e-06,
+ "loss": 0.29573276042938235,
+ "mean_token_accuracy": 0.902192335575819,
+ "num_tokens": 3557030.0,
+ "step": 380
+ },
+ {
+ "entropy": 0.2693369995802641,
+ "epoch": 1.287365813377374,
+ "grad_norm": 1.890625,
+ "learning_rate": 3.974358974358974e-06,
+ "loss": 0.27785425186157225,
+ "mean_token_accuracy": 0.9090337291359901,
+ "num_tokens": 3654159.0,
+ "step": 390
+ },
+ {
+ "entropy": 0.2908114368095994,
+ "epoch": 1.3203963666391412,
+ "grad_norm": 2.015625,
+ "learning_rate": 3.7912087912087915e-06,
+ "loss": 0.30535905361175536,
+ "mean_token_accuracy": 0.901054035872221,
+ "num_tokens": 3749664.0,
+ "step": 400
+ },
+ {
+ "epoch": 1.3203963666391412,
+ "eval_entropy": 0.2783401116176888,
+ "eval_loss": 0.28495287895202637,
+ "eval_mean_token_accuracy": 0.9069356251645971,
+ "eval_num_tokens": 3749664.0,
+ "eval_runtime": 17.8814,
+ "eval_samples_per_second": 30.143,
+ "eval_steps_per_second": 15.099,
+ "step": 400
+ },
+ {
+ "entropy": 0.2749379588291049,
+ "epoch": 1.3534269199009084,
+ "grad_norm": 1.84375,
+ "learning_rate": 3.6080586080586083e-06,
+ "loss": 0.27795455455780027,
+ "mean_token_accuracy": 0.9090839587152004,
+ "num_tokens": 3845841.0,
+ "step": 410
+ },
+ {
+ "entropy": 0.2859474029392004,
+ "epoch": 1.3864574731626755,
+ "grad_norm": 1.6796875,
+ "learning_rate": 3.424908424908425e-06,
+ "loss": 0.2886682987213135,
+ "mean_token_accuracy": 0.9059950634837151,
+ "num_tokens": 3942402.0,
+ "step": 420
+ },
+ {
+ "entropy": 0.2537162031978369,
+ "epoch": 1.4194880264244425,
+ "grad_norm": 1.9375,
+ "learning_rate": 3.2417582417582424e-06,
+ "loss": 0.25533475875854494,
+ "mean_token_accuracy": 0.9158934533596039,
+ "num_tokens": 4034122.0,
+ "step": 430
+ },
+ {
+ "entropy": 0.28255065642297267,
+ "epoch": 1.4525185796862097,
+ "grad_norm": 1.7734375,
+ "learning_rate": 3.058608058608059e-06,
+ "loss": 0.29817631244659426,
+ "mean_token_accuracy": 0.9040707617998123,
+ "num_tokens": 4127068.0,
+ "step": 440
+ },
+ {
+ "entropy": 0.2898306529968977,
+ "epoch": 1.4855491329479769,
+ "grad_norm": 1.703125,
+ "learning_rate": 2.8754578754578755e-06,
+ "loss": 0.2923931121826172,
+ "mean_token_accuracy": 0.9030195340514183,
+ "num_tokens": 4224435.0,
+ "step": 450
+ },
+ {
+ "entropy": 0.27209411058574917,
+ "epoch": 1.518579686209744,
+ "grad_norm": 1.7578125,
+ "learning_rate": 2.6923076923076923e-06,
+ "loss": 0.28646528720855713,
+ "mean_token_accuracy": 0.9095881894230843,
+ "num_tokens": 4321028.0,
+ "step": 460
+ },
+ {
+ "entropy": 0.280727200768888,
+ "epoch": 1.5516102394715112,
+ "grad_norm": 1.7421875,
+ "learning_rate": 2.509157509157509e-06,
+ "loss": 0.2891413688659668,
+ "mean_token_accuracy": 0.9067636162042618,
+ "num_tokens": 4414916.0,
+ "step": 470
+ },
+ {
+ "entropy": 0.27109829243272543,
+ "epoch": 1.5846407927332784,
+ "grad_norm": 1.8359375,
+ "learning_rate": 2.3260073260073264e-06,
+ "loss": 0.26625516414642336,
+ "mean_token_accuracy": 0.9099020563066006,
+ "num_tokens": 4505708.0,
+ "step": 480
+ },
+ {
+ "entropy": 0.2729395367205143,
+ "epoch": 1.6176713459950454,
+ "grad_norm": 1.7890625,
+ "learning_rate": 2.1428571428571427e-06,
+ "loss": 0.2779061317443848,
+ "mean_token_accuracy": 0.909747340530157,
+ "num_tokens": 4600244.0,
+ "step": 490
+ },
+ {
+ "entropy": 0.2682301852852106,
+ "epoch": 1.6507018992568125,
+ "grad_norm": 1.6953125,
+ "learning_rate": 1.95970695970696e-06,
+ "loss": 0.274446964263916,
+ "mean_token_accuracy": 0.9094639919698239,
+ "num_tokens": 4688147.0,
+ "step": 500
+ },
+ {
+ "epoch": 1.6507018992568125,
+ "eval_entropy": 0.2756172451708052,
+ "eval_loss": 0.2841346263885498,
+ "eval_mean_token_accuracy": 0.9072470570052111,
+ "eval_num_tokens": 4688147.0,
+ "eval_runtime": 17.9119,
+ "eval_samples_per_second": 30.092,
+ "eval_steps_per_second": 15.074,
+ "step": 500
+ },
+ {
+ "entropy": 0.26342453975230457,
+ "epoch": 1.6837324525185797,
+ "grad_norm": 1.890625,
+ "learning_rate": 1.7765567765567768e-06,
+ "loss": 0.26432271003723146,
+ "mean_token_accuracy": 0.9114409819245338,
+ "num_tokens": 4780124.0,
+ "step": 510
+ },
+ {
+ "entropy": 0.28657868495211003,
+ "epoch": 1.7167630057803467,
+ "grad_norm": 1.7734375,
+ "learning_rate": 1.5934065934065933e-06,
+ "loss": 0.2983593225479126,
+ "mean_token_accuracy": 0.9028328903019428,
+ "num_tokens": 4874633.0,
+ "step": 520
+ },
+ {
+ "entropy": 0.25633321776986123,
+ "epoch": 1.7497935590421139,
+ "grad_norm": 1.8359375,
+ "learning_rate": 1.4102564102564104e-06,
+ "loss": 0.2606488227844238,
+ "mean_token_accuracy": 0.9153999648988247,
+ "num_tokens": 4967263.0,
+ "step": 530
+ },
+ {
+ "entropy": 0.26120030507445335,
+ "epoch": 1.782824112303881,
+ "grad_norm": 2.078125,
+ "learning_rate": 1.2271062271062271e-06,
+ "loss": 0.2706313610076904,
+ "mean_token_accuracy": 0.9130682416260243,
+ "num_tokens": 5056736.0,
+ "step": 540
+ },
+ {
+ "entropy": 0.27009565513581035,
+ "epoch": 1.8158546655656482,
+ "grad_norm": 1.8828125,
+ "learning_rate": 1.0439560439560442e-06,
+ "loss": 0.277858567237854,
+ "mean_token_accuracy": 0.9087999157607556,
+ "num_tokens": 5150617.0,
+ "step": 550
+ },
+ {
+ "entropy": 0.2819518899545074,
+ "epoch": 1.8488852188274154,
+ "grad_norm": 1.9609375,
+ "learning_rate": 8.608058608058609e-07,
+ "loss": 0.28743035793304444,
+ "mean_token_accuracy": 0.9067244723439216,
+ "num_tokens": 5243538.0,
+ "step": 560
+ },
+ {
+ "entropy": 0.2858045116066933,
+ "epoch": 1.8819157720891826,
+ "grad_norm": 2.203125,
+ "learning_rate": 6.776556776556777e-07,
+ "loss": 0.2939825296401978,
+ "mean_token_accuracy": 0.904579871147871,
+ "num_tokens": 5331441.0,
+ "step": 570
+ },
+ {
+ "entropy": 0.2926233192905784,
+ "epoch": 1.9149463253509498,
+ "grad_norm": 2.3125,
+ "learning_rate": 4.945054945054946e-07,
+ "loss": 0.29961700439453126,
+ "mean_token_accuracy": 0.9033655568957328,
+ "num_tokens": 5426734.0,
+ "step": 580
+ },
+ {
+ "entropy": 0.2706678254529834,
+ "epoch": 1.9479768786127167,
+ "grad_norm": 2.03125,
+ "learning_rate": 3.113553113553114e-07,
+ "loss": 0.27953526973724363,
+ "mean_token_accuracy": 0.9093762949109078,
+ "num_tokens": 5518190.0,
+ "step": 590
+ },
+ {
+ "entropy": 0.27056904807686805,
+ "epoch": 1.981007431874484,
+ "grad_norm": 2.09375,
+ "learning_rate": 1.282051282051282e-07,
+ "loss": 0.2764536619186401,
+ "mean_token_accuracy": 0.9110228553414345,
+ "num_tokens": 5609238.0,
+ "step": 600
+ },
+ {
+ "epoch": 1.981007431874484,
+ "eval_entropy": 0.2757842331572815,
+ "eval_loss": 0.28394654393196106,
+ "eval_mean_token_accuracy": 0.9073928307603907,
+ "eval_num_tokens": 5609238.0,
+ "eval_runtime": 18.1138,
+ "eval_samples_per_second": 29.756,
+ "eval_steps_per_second": 14.906,
+ "step": 600
+ },
+ {
+ "epoch": 2.0,
+ "eval_entropy": 0.2757109238593667,
+ "eval_loss": 0.2840007245540619,
+ "eval_mean_token_accuracy": 0.9072579809912928,
+ "eval_num_tokens": 5663138.0,
+ "eval_runtime": 18.1948,
+ "eval_samples_per_second": 29.624,
+ "eval_steps_per_second": 14.839,
+ "step": 606
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 606,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 2,
+ "save_steps": 100,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": true
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 6.072645019326874e+16,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-606/training_args.bin b/checkpoint-606/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..133cb6661b0124bb9db0d7d208d748fd736ea689
--- /dev/null
+++ b/checkpoint-606/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6696e1fef8bde5e22de2f5ae18c284f2b2d35a345157db5151571518385fb38b
+size 5713
diff --git a/checkpoint-674/chat_template.jinja b/checkpoint-674/chat_template.jinja
new file mode 100644
index 0000000000000000000000000000000000000000..01be9b307daa2d425f7c168c9fb145a286e0afb4
--- /dev/null
+++ b/checkpoint-674/chat_template.jinja
@@ -0,0 +1,89 @@
+{%- if tools %}
+ {{- '<|im_start|>system\n' }}
+ {%- if messages[0].role == 'system' %}
+ {{- messages[0].content + '\n\n' }}
+ {%- endif %}
+ {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within XML tags:\n" }}
+ {%- for tool in tools %}
+ {{- "\n" }}
+ {{- tool | tojson }}
+ {%- endfor %}
+ {{- "\n\n\nFor each function call, return a json object with function name and arguments within XML tags:\n\n{\"name\": , \"arguments\": }\n<|im_end|>\n" }}
+{%- else %}
+ {%- if messages[0].role == 'system' %}
+ {{- '<|im_start|>system\n' + messages[0].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" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}
+ {%- set ns.multi_step_tool = false %}
+ {%- set ns.last_query_index = index %}
+ {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+ {%- if message.content is string %}
+ {%- set content = message.content %}
+ {%- else %}
+ {%- set content = '' %}
+ {%- endif %}
+ {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+ {{- '<|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 '' in content %}
+ {%- set reasoning_content = content.split('')[0].rstrip('\n').split('')[-1].lstrip('\n') %}
+ {%- set content = content.split('')[-1].lstrip('\n') %}
+ {%- endif %}
+ {%- endif %}
+ {%- if loop.index0 > ns.last_query_index %}
+ {%- if loop.last or (not loop.last and reasoning_content) %}
+ {{- '<|im_start|>' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\n\n' + content.lstrip('\n') }}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- else %}
+ {{- '<|im_start|>' + message.role + '\n' + content }}
+ {%- endif %}
+ {%- if message.tool_calls %}
+ {%- for tool_call in message.tool_calls %}
+ {%- if (loop.first and content) or (not loop.first) %}
+ {{- '\n' }}
+ {%- endif %}
+ {%- if tool_call.function %}
+ {%- set tool_call = tool_call.function %}
+ {%- endif %}
+ {{- '\n{"name": "' }}
+ {{- tool_call.name }}
+ {{- '", "arguments": ' }}
+ {%- if tool_call.arguments is string %}
+ {{- tool_call.arguments }}
+ {%- else %}
+ {{- tool_call.arguments | tojson }}
+ {%- endif %}
+ {{- '}\n' }}
+ {%- endfor %}
+ {%- endif %}
+ {{- '<|im_end|>\n' }}
+ {%- elif message.role == "tool" %}
+ {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+ {{- '<|im_start|>user' }}
+ {%- endif %}
+ {{- '\n\n' }}
+ {{- content }}
+ {{- '\n' }}
+ {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+ {{- '<|im_end|>\n' }}
+ {%- endif %}
+ {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+ {{- '<|im_start|>assistant\n' }}
+ {%- if enable_thinking is defined and enable_thinking is false %}
+ {{- '\n\n\n\n' }}
+ {%- endif %}
+{%- endif %}
\ No newline at end of file
diff --git a/checkpoint-674/config.json b/checkpoint-674/config.json
new file mode 100644
index 0000000000000000000000000000000000000000..1d219d6a6d288625229a71646710b85cf7ae5def
--- /dev/null
+++ b/checkpoint-674/config.json
@@ -0,0 +1,63 @@
+{
+ "architectures": [
+ "Qwen3ForCausalLM"
+ ],
+ "attention_bias": false,
+ "attention_dropout": 0.0,
+ "bos_token_id": null,
+ "dtype": "bfloat16",
+ "eos_token_id": 151645,
+ "head_dim": 128,
+ "hidden_act": "silu",
+ "hidden_size": 2048,
+ "initializer_range": 0.02,
+ "intermediate_size": 6144,
+ "layer_types": [
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention",
+ "full_attention"
+ ],
+ "max_position_embeddings": 40960,
+ "max_window_layers": 28,
+ "model_type": "qwen3",
+ "num_attention_heads": 16,
+ "num_hidden_layers": 28,
+ "num_key_value_heads": 8,
+ "pad_token_id": 151643,
+ "rms_norm_eps": 1e-06,
+ "rope_parameters": {
+ "rope_theta": 1000000,
+ "rope_type": "default"
+ },
+ "sliding_window": null,
+ "tie_word_embeddings": true,
+ "transformers_version": "5.7.0",
+ "use_cache": false,
+ "use_sliding_window": false,
+ "vocab_size": 151936
+}
diff --git a/checkpoint-674/generation_config.json b/checkpoint-674/generation_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..5ec133d867b1427d43a5ee0ea332bd088fc12013
--- /dev/null
+++ b/checkpoint-674/generation_config.json
@@ -0,0 +1,12 @@
+{
+ "do_sample": true,
+ "eos_token_id": [
+ 151645,
+ 151643
+ ],
+ "pad_token_id": 151643,
+ "temperature": 0.6,
+ "top_k": 20,
+ "top_p": 0.95,
+ "transformers_version": "5.7.0"
+}
diff --git a/checkpoint-674/model.safetensors b/checkpoint-674/model.safetensors
new file mode 100644
index 0000000000000000000000000000000000000000..cc885a501a0fffef97b843b5248e3bb7cfe46172
--- /dev/null
+++ b/checkpoint-674/model.safetensors
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3af4557502ae7c7853ed3b684aa50cfd00317aafc237f2dda688075bb3dc81c1
+size 3441185608
diff --git a/checkpoint-674/optimizer.pt b/checkpoint-674/optimizer.pt
new file mode 100644
index 0000000000000000000000000000000000000000..4f8c22d5c745517a99843c614c05da8a366a7d78
--- /dev/null
+++ b/checkpoint-674/optimizer.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ca0ddbeff189519bd91d1474687c97eb42f33e1dc183a5769974dadc5e0cd7b
+size 6882572207
diff --git a/checkpoint-674/rng_state.pth b/checkpoint-674/rng_state.pth
new file mode 100644
index 0000000000000000000000000000000000000000..1feba1a6538e93b94696d3773853dbc8947b0cad
--- /dev/null
+++ b/checkpoint-674/rng_state.pth
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:098b29492211804ab324a36f37466821d948280bb74fce4ba895c03f13ecd878
+size 14645
diff --git a/checkpoint-674/scheduler.pt b/checkpoint-674/scheduler.pt
new file mode 100644
index 0000000000000000000000000000000000000000..52e04de5969d9a926e44ed315d79199b8ca27698
--- /dev/null
+++ b/checkpoint-674/scheduler.pt
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01dbf241e935857d2a5486058c590aafb795b79d6ef3f0dc038c0d85e2187d7d
+size 1465
diff --git a/checkpoint-674/tokenizer.json b/checkpoint-674/tokenizer.json
new file mode 100644
index 0000000000000000000000000000000000000000..c7afbed2efcdf019f88ab0572ec29d3bf595dfe2
--- /dev/null
+++ b/checkpoint-674/tokenizer.json
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be75606093db2094d7cd20f3c2f385c212750648bd6ea4fb2bf507a6a4c55506
+size 11422650
diff --git a/checkpoint-674/tokenizer_config.json b/checkpoint-674/tokenizer_config.json
new file mode 100644
index 0000000000000000000000000000000000000000..770e41d6c92519d525eede4cbcf3ba27f6425311
--- /dev/null
+++ b/checkpoint-674/tokenizer_config.json
@@ -0,0 +1,30 @@
+{
+ "add_prefix_space": false,
+ "backend": "tokenizers",
+ "bos_token": null,
+ "clean_up_tokenization_spaces": false,
+ "eos_token": "<|im_end|>",
+ "errors": "replace",
+ "extra_special_tokens": [
+ "<|im_start|>",
+ "<|im_end|>",
+ "<|object_ref_start|>",
+ "<|object_ref_end|>",
+ "<|box_start|>",
+ "<|box_end|>",
+ "<|quad_start|>",
+ "<|quad_end|>",
+ "<|vision_start|>",
+ "<|vision_end|>",
+ "<|vision_pad|>",
+ "<|image_pad|>",
+ "<|video_pad|>"
+ ],
+ "is_local": false,
+ "local_files_only": false,
+ "model_max_length": 131072,
+ "pad_token": "<|endoftext|>",
+ "split_special_tokens": false,
+ "tokenizer_class": "Qwen2Tokenizer",
+ "unk_token": null
+}
diff --git a/checkpoint-674/trainer_state.json b/checkpoint-674/trainer_state.json
new file mode 100644
index 0000000000000000000000000000000000000000..54b5d973ad7d8fb6a649dbc5f72a0c5653d31516
--- /dev/null
+++ b/checkpoint-674/trainer_state.json
@@ -0,0 +1,704 @@
+{
+ "best_global_step": null,
+ "best_metric": null,
+ "best_model_checkpoint": null,
+ "epoch": 2.0,
+ "eval_steps": 500,
+ "global_step": 674,
+ "is_hyper_param_search": false,
+ "is_local_process_zero": true,
+ "is_world_process_zero": true,
+ "log_history": [
+ {
+ "entropy": NaN,
+ "epoch": 0.029717682020802376,
+ "grad_norm": NaN,
+ "learning_rate": 1.8000000000000001e-06,
+ "loss": 1005514863411.2,
+ "mean_token_accuracy": 0.07770282882993343,
+ "num_tokens": 97764.0,
+ "step": 10
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.05943536404160475,
+ "grad_norm": NaN,
+ "learning_rate": 3.8000000000000005e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007392044077278115,
+ "num_tokens": 190372.0,
+ "step": 20
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.08915304606240713,
+ "grad_norm": NaN,
+ "learning_rate": 5.8e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000383764869911829,
+ "num_tokens": 280928.0,
+ "step": 30
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.1188707280832095,
+ "grad_norm": NaN,
+ "learning_rate": 7.800000000000002e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003212536917999387,
+ "num_tokens": 380644.0,
+ "step": 40
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.1485884101040119,
+ "grad_norm": NaN,
+ "learning_rate": 9.800000000000001e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003853117574180942,
+ "num_tokens": 475467.0,
+ "step": 50
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.17830609212481427,
+ "grad_norm": NaN,
+ "learning_rate": 1.18e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003134153150313068,
+ "num_tokens": 572411.0,
+ "step": 60
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.20802377414561665,
+ "grad_norm": NaN,
+ "learning_rate": 1.38e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011001345737895463,
+ "num_tokens": 661517.0,
+ "step": 70
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.237741456166419,
+ "grad_norm": NaN,
+ "learning_rate": 1.58e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012231820757733658,
+ "num_tokens": 752142.0,
+ "step": 80
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.2674591381872214,
+ "grad_norm": NaN,
+ "learning_rate": 1.7800000000000002e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011551132018212229,
+ "num_tokens": 850399.0,
+ "step": 90
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.2971768202080238,
+ "grad_norm": NaN,
+ "learning_rate": 1.98e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00047943509634933433,
+ "num_tokens": 954759.0,
+ "step": 100
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.32689450222882616,
+ "grad_norm": NaN,
+ "learning_rate": 1.9802414928649837e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008438902725174558,
+ "num_tokens": 1050439.0,
+ "step": 110
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.35661218424962854,
+ "grad_norm": NaN,
+ "learning_rate": 1.958287596048299e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009903057703922967,
+ "num_tokens": 1135089.0,
+ "step": 120
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.3863298662704309,
+ "grad_norm": NaN,
+ "learning_rate": 1.9363336992316137e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00040032515971688554,
+ "num_tokens": 1228149.0,
+ "step": 130
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.4160475482912333,
+ "grad_norm": NaN,
+ "learning_rate": 1.9143798024149288e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0010247966441966128,
+ "num_tokens": 1331672.0,
+ "step": 140
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.4457652303120357,
+ "grad_norm": NaN,
+ "learning_rate": 1.892425905598244e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000843161555530969,
+ "num_tokens": 1421314.0,
+ "step": 150
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.475482912332838,
+ "grad_norm": NaN,
+ "learning_rate": 1.8704720087815587e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007889680869993754,
+ "num_tokens": 1511701.0,
+ "step": 160
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5052005943536404,
+ "grad_norm": NaN,
+ "learning_rate": 1.848518111964874e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007493881763366517,
+ "num_tokens": 1605063.0,
+ "step": 170
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5349182763744428,
+ "grad_norm": NaN,
+ "learning_rate": 1.826564215148189e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000567350817436818,
+ "num_tokens": 1703107.0,
+ "step": 180
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5646359583952452,
+ "grad_norm": NaN,
+ "learning_rate": 1.804610318331504e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00027060438151238486,
+ "num_tokens": 1796279.0,
+ "step": 190
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.5943536404160475,
+ "grad_norm": NaN,
+ "learning_rate": 1.782656421514819e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008226798396208324,
+ "num_tokens": 1893647.0,
+ "step": 200
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6240713224368499,
+ "grad_norm": NaN,
+ "learning_rate": 1.760702524698134e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007338574505411088,
+ "num_tokens": 1982325.0,
+ "step": 210
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6537890044576523,
+ "grad_norm": NaN,
+ "learning_rate": 1.7387486278814492e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004891513846814633,
+ "num_tokens": 2071869.0,
+ "step": 220
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.6835066864784547,
+ "grad_norm": NaN,
+ "learning_rate": 1.716794731064764e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00038509477453771976,
+ "num_tokens": 2168370.0,
+ "step": 230
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7132243684992571,
+ "grad_norm": NaN,
+ "learning_rate": 1.694840834248079e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009255842014681548,
+ "num_tokens": 2263211.0,
+ "step": 240
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7429420505200595,
+ "grad_norm": NaN,
+ "learning_rate": 1.6728869374313943e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011857455348945223,
+ "num_tokens": 2352425.0,
+ "step": 250
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.7726597325408618,
+ "grad_norm": NaN,
+ "learning_rate": 1.6509330406147094e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008204675337765366,
+ "num_tokens": 2445217.0,
+ "step": 260
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8023774145616642,
+ "grad_norm": NaN,
+ "learning_rate": 1.6289791437980242e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0013553448639868292,
+ "num_tokens": 2537677.0,
+ "step": 270
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8320950965824666,
+ "grad_norm": NaN,
+ "learning_rate": 1.6070252469813393e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011630529144895263,
+ "num_tokens": 2631646.0,
+ "step": 280
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.861812778603269,
+ "grad_norm": NaN,
+ "learning_rate": 1.5850713501646545e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009409735575900413,
+ "num_tokens": 2722632.0,
+ "step": 290
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.8915304606240714,
+ "grad_norm": NaN,
+ "learning_rate": 1.5631174533479693e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006484775040007662,
+ "num_tokens": 2819185.0,
+ "step": 300
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.9212481426448736,
+ "grad_norm": NaN,
+ "learning_rate": 1.5411635565312844e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011682231881422922,
+ "num_tokens": 2910483.0,
+ "step": 310
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.950965824665676,
+ "grad_norm": NaN,
+ "learning_rate": 1.5192096597145995e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004708930806373246,
+ "num_tokens": 2997057.0,
+ "step": 320
+ },
+ {
+ "entropy": NaN,
+ "epoch": 0.9806835066864784,
+ "grad_norm": NaN,
+ "learning_rate": 1.4972557628979145e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011955970534472727,
+ "num_tokens": 3091284.0,
+ "step": 330
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0089153046062407,
+ "grad_norm": NaN,
+ "learning_rate": 1.4753018660812295e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006843029964090276,
+ "num_tokens": 3177800.0,
+ "step": 340
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0386329866270432,
+ "grad_norm": NaN,
+ "learning_rate": 1.4533479692645444e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0005406756106822285,
+ "num_tokens": 3271989.0,
+ "step": 350
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.0683506686478454,
+ "grad_norm": NaN,
+ "learning_rate": 1.4313940724478596e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007009803419350646,
+ "num_tokens": 3365082.0,
+ "step": 360
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.098068350668648,
+ "grad_norm": NaN,
+ "learning_rate": 1.4094401756311746e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001003894711175235,
+ "num_tokens": 3465865.0,
+ "step": 370
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.1277860326894502,
+ "grad_norm": NaN,
+ "learning_rate": 1.3874862788144897e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004136150790145621,
+ "num_tokens": 3564998.0,
+ "step": 380
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.1575037147102525,
+ "grad_norm": NaN,
+ "learning_rate": 1.3655323819978048e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000998112405068241,
+ "num_tokens": 3661508.0,
+ "step": 390
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.187221396731055,
+ "grad_norm": NaN,
+ "learning_rate": 1.3435784851811198e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009621107514249161,
+ "num_tokens": 3755395.0,
+ "step": 400
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.2169390787518575,
+ "grad_norm": NaN,
+ "learning_rate": 1.3216245883644348e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004976072828867473,
+ "num_tokens": 3842994.0,
+ "step": 410
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.2466567607726597,
+ "grad_norm": NaN,
+ "learning_rate": 1.2996706915477499e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0003434549740632065,
+ "num_tokens": 3934618.0,
+ "step": 420
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.276374442793462,
+ "grad_norm": NaN,
+ "learning_rate": 1.2777167947310649e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00026364591685705817,
+ "num_tokens": 4026711.0,
+ "step": 430
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3060921248142645,
+ "grad_norm": NaN,
+ "learning_rate": 1.2557628979143798e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006349188319290988,
+ "num_tokens": 4114609.0,
+ "step": 440
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3358098068350668,
+ "grad_norm": NaN,
+ "learning_rate": 1.2338090010976948e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004265022376785055,
+ "num_tokens": 4203953.0,
+ "step": 450
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3655274888558693,
+ "grad_norm": NaN,
+ "learning_rate": 1.2118551042810101e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012633154692593962,
+ "num_tokens": 4298319.0,
+ "step": 460
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.3952451708766715,
+ "grad_norm": NaN,
+ "learning_rate": 1.189901207464325e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0011256658195634372,
+ "num_tokens": 4396020.0,
+ "step": 470
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.424962852897474,
+ "grad_norm": NaN,
+ "learning_rate": 1.16794731064764e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0013131745137798135,
+ "num_tokens": 4491203.0,
+ "step": 480
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.4546805349182763,
+ "grad_norm": NaN,
+ "learning_rate": 1.1459934138309552e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012716218378045596,
+ "num_tokens": 4582199.0,
+ "step": 490
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.4843982169390788,
+ "grad_norm": NaN,
+ "learning_rate": 1.1240395170142701e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0014869970778818241,
+ "num_tokens": 4672589.0,
+ "step": 500
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.514115898959881,
+ "grad_norm": NaN,
+ "learning_rate": 1.1020856201975851e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00046981508567114363,
+ "num_tokens": 4767693.0,
+ "step": 510
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.5438335809806834,
+ "grad_norm": NaN,
+ "learning_rate": 1.0801317233809e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000601445968641201,
+ "num_tokens": 4860371.0,
+ "step": 520
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.5735512630014858,
+ "grad_norm": NaN,
+ "learning_rate": 1.0581778265642154e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012073652920662425,
+ "num_tokens": 4950622.0,
+ "step": 530
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6032689450222883,
+ "grad_norm": NaN,
+ "learning_rate": 1.0362239297475303e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.00028707647215924225,
+ "num_tokens": 5044878.0,
+ "step": 540
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6329866270430906,
+ "grad_norm": NaN,
+ "learning_rate": 1.0142700329308453e-05,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004101752907445189,
+ "num_tokens": 5140938.0,
+ "step": 550
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6627043090638929,
+ "grad_norm": NaN,
+ "learning_rate": 9.923161361141603e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008072987395280507,
+ "num_tokens": 5232852.0,
+ "step": 560
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.6924219910846954,
+ "grad_norm": NaN,
+ "learning_rate": 9.703622392974754e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0012222853780258446,
+ "num_tokens": 5321835.0,
+ "step": 570
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7221396731054979,
+ "grad_norm": NaN,
+ "learning_rate": 9.484083424807904e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001064640874392353,
+ "num_tokens": 5416464.0,
+ "step": 580
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7518573551263001,
+ "grad_norm": NaN,
+ "learning_rate": 9.264544456641055e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0009612867230316624,
+ "num_tokens": 5515169.0,
+ "step": 590
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.7815750371471024,
+ "grad_norm": NaN,
+ "learning_rate": 9.045005488474205e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008774219051701948,
+ "num_tokens": 5611971.0,
+ "step": 600
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.811292719167905,
+ "grad_norm": NaN,
+ "learning_rate": 8.825466520307355e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007863394406740553,
+ "num_tokens": 5704255.0,
+ "step": 610
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.8410104011887074,
+ "grad_norm": NaN,
+ "learning_rate": 8.605927552140506e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.000578320066415472,
+ "num_tokens": 5800440.0,
+ "step": 620
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.8707280832095097,
+ "grad_norm": NaN,
+ "learning_rate": 8.386388583973656e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0004275172817870043,
+ "num_tokens": 5890898.0,
+ "step": 630
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.900445765230312,
+ "grad_norm": NaN,
+ "learning_rate": 8.166849615806807e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.001425291335908696,
+ "num_tokens": 5987228.0,
+ "step": 640
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.9301634472511144,
+ "grad_norm": NaN,
+ "learning_rate": 7.947310647639957e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0006059414023184218,
+ "num_tokens": 6076579.0,
+ "step": 650
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.959881129271917,
+ "grad_norm": NaN,
+ "learning_rate": 7.727771679473108e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0008722793892957271,
+ "num_tokens": 6171013.0,
+ "step": 660
+ },
+ {
+ "entropy": NaN,
+ "epoch": 1.9895988112927192,
+ "grad_norm": NaN,
+ "learning_rate": 7.508232711306257e-06,
+ "loss": 0.0,
+ "mean_token_accuracy": 0.0007748565993097145,
+ "num_tokens": 6265432.0,
+ "step": 670
+ }
+ ],
+ "logging_steps": 10,
+ "max_steps": 1011,
+ "num_input_tokens_seen": 0,
+ "num_train_epochs": 3,
+ "save_steps": 500,
+ "stateful_callbacks": {
+ "TrainerControl": {
+ "args": {
+ "should_epoch_stop": false,
+ "should_evaluate": false,
+ "should_log": false,
+ "should_save": true,
+ "should_training_stop": false
+ },
+ "attributes": {}
+ }
+ },
+ "total_flos": 6.753063653143757e+16,
+ "train_batch_size": 2,
+ "trial_name": null,
+ "trial_params": null
+}
diff --git a/checkpoint-674/training_args.bin b/checkpoint-674/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..737a87caa4fc2f704a1fe56a508ca4d335ca783d
--- /dev/null
+++ b/checkpoint-674/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae0f7084660199ed7fd16027c0c6af5e726790f32a022b019f1277bec4e01142
+size 5713
diff --git a/eval_results.diverse.json b/eval_results.diverse.json
new file mode 100644
index 0000000000000000000000000000000000000000..1f6ec55ddc2f6134480996e20074310f889d6b16
--- /dev/null
+++ b/eval_results.diverse.json
@@ -0,0 +1,207 @@
+{
+ "checkpoint": "/scratch/checkpoints/exp6_sft_numinamath_dpo",
+ "is_local": true,
+ "eval_tier": "diverse",
+ "sampling": {
+ "n": 8,
+ "temperature": 0.6,
+ "top_p": 0.95,
+ "top_k": 20,
+ "max_tokens": 3500
+ },
+ "benchmarks": {
+ "numinamath_benchmark": {
+ "source": "numinamath_benchmark",
+ "overall": {
+ "n_problems": 500,
+ "n_completions": 8,
+ "pass@1": 0.3795,
+ "pass@8": 0.612
+ },
+ "by_source": {
+ "numina:amc_aime": {
+ "n_problems": 1,
+ "n_completions": 8,
+ "pass@1": 0.25,
+ "pass@8": 1.0
+ },
+ "numina:aops_forum": {
+ "n_problems": 9,
+ "n_completions": 8,
+ "pass@1": 0.08333333333333333,
+ "pass@8": 0.4444444444444444
+ },
+ "numina:cn_contest": {
+ "n_problems": 7,
+ "n_completions": 8,
+ "pass@1": 0.0,
+ "pass@8": 0.0
+ },
+ "numina:cn_k12": {
+ "n_problems": 200,
+ "n_completions": 8,
+ "pass@1": 0.331875,
+ "pass@8": 0.62
+ },
+ "numina:inequalities": {
+ "n_problems": 1,
+ "n_completions": 8,
+ "pass@1": 0.0,
+ "pass@8": 0.0
+ },
+ "numina:metamath": {
+ "n_problems": 7,
+ "n_completions": 8,
+ "pass@1": 0.1964285714285714,
+ "pass@8": 0.7142857142857143
+ },
+ "numina:olympiads": {
+ "n_problems": 49,
+ "n_completions": 8,
+ "pass@1": 0.06887755102040816,
+ "pass@8": 0.20408163265306123
+ },
+ "numina:olympiads_ref": {
+ "n_problems": 1,
+ "n_completions": 8,
+ "pass@1": 0.25,
+ "pass@8": 1.0
+ },
+ "numina:orca_math": {
+ "n_problems": 112,
+ "n_completions": 8,
+ "pass@1": 0.6149553571428571,
+ "pass@8": 0.8125
+ },
+ "numina:synthetic_math": {
+ "n_problems": 113,
+ "n_completions": 8,
+ "pass@1": 0.42920353982300885,
+ "pass@8": 0.6194690265486725
+ }
+ },
+ "by_topic": {
+ "Algebra": {
+ "n_problems": 277,
+ "n_completions": 8,
+ "pass@1": 0.4453971119133574,
+ "pass@8": 0.6895306859205776
+ },
+ "Calculus": {
+ "n_problems": 17,
+ "n_completions": 8,
+ "pass@1": 0.20588235294117646,
+ "pass@8": 0.35294117647058826
+ },
+ "Combinatorics": {
+ "n_problems": 31,
+ "n_completions": 8,
+ "pass@1": 0.2862903225806452,
+ "pass@8": 0.5806451612903226
+ },
+ "Geometry": {
+ "n_problems": 84,
+ "n_completions": 8,
+ "pass@1": 0.3169642857142857,
+ "pass@8": 0.5357142857142857
+ },
+ "Inequalities": {
+ "n_problems": 11,
+ "n_completions": 8,
+ "pass@1": 0.375,
+ "pass@8": 0.6363636363636364
+ },
+ "Logic and Puzzles": {
+ "n_problems": 20,
+ "n_completions": 8,
+ "pass@1": 0.45,
+ "pass@8": 0.65
+ },
+ "Number Theory": {
+ "n_problems": 54,
+ "n_completions": 8,
+ "pass@1": 0.24537037037037038,
+ "pass@8": 0.46296296296296297
+ },
+ "Other": {
+ "n_problems": 6,
+ "n_completions": 8,
+ "pass@1": 0.16666666666666666,
+ "pass@8": 0.16666666666666666
+ }
+ }
+ },
+ "diverse_math_benchmark": {
+ "source": "diverse_math_benchmark",
+ "overall": {
+ "n_problems": 150,
+ "n_completions": 8,
+ "pass@1": 0.3275,
+ "pass@8": 0.5466666666666666
+ },
+ "by_source": {
+ "aime_2024": {
+ "n_problems": 17,
+ "n_completions": 8,
+ "pass@1": 0.036764705882352935,
+ "pass@8": 0.11764705882352941
+ },
+ "aime_2025": {
+ "n_problems": 13,
+ "n_completions": 8,
+ "pass@1": 0.08653846153846154,
+ "pass@8": 0.15384615384615385
+ },
+ "math_hard": {
+ "n_problems": 50,
+ "n_completions": 8,
+ "pass@1": 0.495,
+ "pass@8": 0.78
+ },
+ "numina:amc_aime": {
+ "n_problems": 1,
+ "n_completions": 8,
+ "pass@1": 0.125,
+ "pass@8": 1.0
+ },
+ "numina:aops_forum": {
+ "n_problems": 2,
+ "n_completions": 8,
+ "pass@1": 0.0,
+ "pass@8": 0.0
+ },
+ "numina:cn_k12": {
+ "n_problems": 26,
+ "n_completions": 8,
+ "pass@1": 0.32211538461538464,
+ "pass@8": 0.5769230769230769
+ },
+ "numina:metamath": {
+ "n_problems": 1,
+ "n_completions": 8,
+ "pass@1": 0.0,
+ "pass@8": 0.0
+ },
+ "numina:olympiads": {
+ "n_problems": 17,
+ "n_completions": 8,
+ "pass@1": 0.08088235294117647,
+ "pass@8": 0.29411764705882354
+ },
+ "numina:orca_math": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.6964285714285714,
+ "pass@8": 0.9285714285714286
+ },
+ "numina:synthetic_math": {
+ "n_problems": 9,
+ "n_completions": 8,
+ "pass@1": 0.3333333333333333,
+ "pass@8": 0.5555555555555556
+ }
+ }
+ }
+ },
+ "note": "All benchmarks use identical sampling params so numbers compare directly across experiments (baseline, SFT v1/v2, GRPO). Hard benchmarks (math_hard_100, aime_benchmark) are decontaminated against MixLong (the SFT pool)."
+}
\ No newline at end of file
diff --git a/eval_results.json b/eval_results.json
new file mode 100644
index 0000000000000000000000000000000000000000..34bf083a93ff87791f872d45ce86fa8c43f28737
--- /dev/null
+++ b/eval_results.json
@@ -0,0 +1,114 @@
+{
+ "checkpoint": "/scratch/checkpoints/exp6_sft_numinamath_dpo",
+ "is_local": true,
+ "eval_tier": "fast",
+ "sampling": {
+ "n": 8,
+ "temperature": 0.3,
+ "top_p": 0.95,
+ "top_k": 20,
+ "max_tokens": 3500
+ },
+ "benchmarks": {
+ "ci_gate_math": {
+ "source": "ci_gate_math",
+ "overall": {
+ "n_problems": 10,
+ "n_completions": 8,
+ "pass@1": 0.2,
+ "pass@8": 0.3
+ }
+ },
+ "math_hard_100": {
+ "source": "math_hard_100",
+ "overall": {
+ "n_problems": 100,
+ "n_completions": 8,
+ "pass@1": 0.49375,
+ "pass@8": 0.75
+ },
+ "by_level": {
+ "4": {
+ "n_problems": 51,
+ "n_completions": 8,
+ "pass@1": 0.6029411764705882,
+ "pass@8": 0.8431372549019608
+ },
+ "5": {
+ "n_problems": 49,
+ "n_completions": 8,
+ "pass@1": 0.38010204081632654,
+ "pass@8": 0.6530612244897959
+ }
+ },
+ "by_topic": {
+ "algebra": {
+ "n_problems": 15,
+ "n_completions": 8,
+ "pass@1": 0.7916666666666666,
+ "pass@8": 0.9333333333333333
+ },
+ "counting_and_probability": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.2857142857142857,
+ "pass@8": 0.5
+ },
+ "geometry": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.3928571428571428,
+ "pass@8": 0.7142857142857143
+ },
+ "intermediate_algebra": {
+ "n_problems": 15,
+ "n_completions": 8,
+ "pass@1": 0.4166666666666667,
+ "pass@8": 0.6666666666666666
+ },
+ "number_theory": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.5178571428571429,
+ "pass@8": 0.7857142857142857
+ },
+ "prealgebra": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.6607142857142857,
+ "pass@8": 1.0
+ },
+ "precalculus": {
+ "n_problems": 14,
+ "n_completions": 8,
+ "pass@1": 0.375,
+ "pass@8": 0.6428571428571429
+ }
+ }
+ },
+ "aime_benchmark": {
+ "source": "aime_benchmark",
+ "overall": {
+ "n_problems": 60,
+ "n_completions": 8,
+ "pass@1": 0.031249999999999993,
+ "pass@8": 0.11666666666666667
+ },
+ "by_source": {
+ "aime_2024": {
+ "n_problems": 30,
+ "n_completions": 8,
+ "pass@1": 0.029166666666666664,
+ "pass@8": 0.13333333333333333
+ },
+ "aime_2025": {
+ "n_problems": 30,
+ "n_completions": 8,
+ "pass@1": 0.033333333333333326,
+ "pass@8": 0.1
+ }
+ }
+ }
+ },
+ "note": "All benchmarks use identical sampling params so numbers compare directly across experiments (baseline, SFT v1/v2, GRPO). Hard benchmarks (math_hard_100, aime_benchmark) are decontaminated against MixLong (the SFT pool)."
+}
\ No newline at end of file
diff --git a/generation_config.json b/generation_config.json
index 9526fbb9cfca6c6b8911fa4f1995a5c43139fc63..06c1fde472f6fa7f3dcff61cd28741606145fbb3 100644
--- a/generation_config.json
+++ b/generation_config.json
@@ -6,9 +6,8 @@
151643
],
"pad_token_id": 151643,
- "temperature": 0.7,
- "top_p": 0.95,
+ "temperature": 0.3,
"top_k": 20,
- "repetition_penalty": 1.03,
+ "top_p": 0.95,
"transformers_version": "4.51.0"
}
\ No newline at end of file
diff --git a/model.safetensors b/model.safetensors
index a9923328aed6324f3b8bc4b96ecf4cc43763ec69..11f5322c32bff22abb8e5d700808a85101d55fa8 100644
--- a/model.safetensors
+++ b/model.safetensors
@@ -1,3 +1,3 @@
version https://git-lfs.github.com/spec/v1
-oid sha256:c9179fe220727db304cb809795bd582e6dbc90a763ef3579708273b8b09858e5
+oid sha256:30211e80f9e7b89e757cb804386bab11dbe114c7efb2e47b27ee6f86dff63658
size 3441185608
diff --git a/split_info.json b/split_info.json
new file mode 100644
index 0000000000000000000000000000000000000000..8b1b0075d0866579f1bce80b961a23921c0f0863
--- /dev/null
+++ b/split_info.json
@@ -0,0 +1,28 @@
+{
+ "experiment_name": "exp6_sft_numinamath_dpo",
+ "config_file": "/scratch/jennifer/standard-project-m2-the-transformers/configs/exp6_dpo.yaml",
+ "resolved_config": {
+ "num_train_epochs": 1,
+ "learning_rate": 5e-07,
+ "per_device_train_batch_size": 1,
+ "per_device_eval_batch_size": 1,
+ "gradient_accumulation_steps": 8,
+ "warmup_ratio": 0.1,
+ "seed": 42,
+ "save_total_limit": 3,
+ "eval_steps": 50,
+ "logging_steps": 10,
+ "beta": 0.1,
+ "loss_type": "sigmoid",
+ "max_length": 4096,
+ "max_completion_length": 3072,
+ "val_fraction": 0.1,
+ "wandb_project": "cs552-math-dpo",
+ "experiment_name": "exp6_sft_numinamath_dpo",
+ "output_dir": "/scratch/checkpoints/exp6_sft_numinamath_dpo",
+ "sft_checkpoint": "/scratch/checkpoints/exp6_sft_numinamath",
+ "train_source": "/scratch/data/dpo_pairs_v2/pairs.jsonl"
+ },
+ "n_train": 5048,
+ "n_val": 561
+}
\ No newline at end of file
diff --git a/tokenizer_config.json b/tokenizer_config.json
index 480b16d724b41320e1e217f2408c04a8a8e326fc..b0ac7bff8c357331648a433d9b6ae4c1f1381380 100644
--- a/tokenizer_config.json
+++ b/tokenizer_config.json
@@ -27,5 +27,5 @@
"split_special_tokens": false,
"tokenizer_class": "Qwen2Tokenizer",
"unk_token": null,
- "chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {{- messages[0].content + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within XML tags:\\n\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n\\n\\nFor each function call, return a json object with function name and arguments within XML tags:\\n\\n{\\\"name\\\": , \\\"arguments\\\": }\\n<|im_end|>\\n\" }}\n{%- else %}\n {#- BOXING INSTRUCTION injected so it is present even when the caller (e.g. CI)\n passes no system message. Targets our no-box / multi-box failure mode. #}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {{- messages[0].content + '\\n\\n' }}\n {%- endif %}\n {{- 'When solving a math problem, work through it step by step and write your final answer inside a \\boxed{...}.<|im_end|>\\n' }}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if message.content is string %}\n {%- set content = message.content %}\n {%- else %}\n {%- set content = '' %}\n {%- endif %}\n {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- else %}\n {%- if '' in content %}\n {%- set reasoning_content = content.split('')[0].rstrip('\\n').split('')[-1].lstrip('\\n') %}\n {%- set content = content.split('')[-1].lstrip('\\n') %}\n {%- endif %}\n {%- endif %}\n {%- if loop.index0 > ns.last_query_index %}\n {%- if loop.last or (not loop.last and reasoning_content) %}\n {{- '<|im_start|>' + message.role + '\\n\\n' + reasoning_content.strip('\\n') + '\\n\\n\\n' + content.lstrip('\\n') }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n\\n' }}\n {{- content }}\n {{- '\\n' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n {%- if enable_thinking is defined and enable_thinking is false %}\n {{- '\\n\\n\\n\\n' }}\n {%- endif %}\n{%- endif %}"
+ "chat_template": "{%- if tools %}\n {{- '<|im_start|>system\\n' }}\n {%- if messages[0].role == 'system' %}\n {{- messages[0].content + '\\n\\n' }}\n {%- endif %}\n {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within XML tags:\\n\" }}\n {%- for tool in tools %}\n {{- \"\\n\" }}\n {{- tool | tojson }}\n {%- endfor %}\n {{- \"\\n\\n\\nFor each function call, return a json object with function name and arguments within XML tags:\\n\\n{\\\"name\\\": , \\\"arguments\\\": }\\n<|im_end|>\\n\" }}\n{%- else %}\n {%- if messages[0].role == 'system' %}\n {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n {%- set index = (messages|length - 1) - loop.index0 %}\n {%- if ns.multi_step_tool and message.role == \"user\" and message.content is string and not(message.content.startswith('') and message.content.endswith('')) %}\n {%- set ns.multi_step_tool = false %}\n {%- set ns.last_query_index = index %}\n {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n {%- if message.content is string %}\n {%- set content = message.content %}\n {%- else %}\n {%- set content = '' %}\n {%- endif %}\n {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n {{- '<|im_start|>' + message.role + '\\n' + content + '<|im_end|>' + '\\n' }}\n {%- elif message.role == \"assistant\" %}\n {%- set reasoning_content = '' %}\n {%- if message.reasoning_content is string %}\n {%- set reasoning_content = message.reasoning_content %}\n {%- else %}\n {%- if '' in content %}\n {%- set reasoning_content = content.split('')[0].rstrip('\\n').split('')[-1].lstrip('\\n') %}\n {%- set content = content.split('')[-1].lstrip('\\n') %}\n {%- endif %}\n {%- endif %}\n {%- if loop.index0 > ns.last_query_index %}\n {%- if loop.last or (not loop.last and reasoning_content) %}\n {{- '<|im_start|>' + message.role + '\\n\\n' + reasoning_content.strip('\\n') + '\\n\\n\\n' + content.lstrip('\\n') }}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- else %}\n {{- '<|im_start|>' + message.role + '\\n' + content }}\n {%- endif %}\n {%- if message.tool_calls %}\n {%- for tool_call in message.tool_calls %}\n {%- if (loop.first and content) or (not loop.first) %}\n {{- '\\n' }}\n {%- endif %}\n {%- if tool_call.function %}\n {%- set tool_call = tool_call.function %}\n {%- endif %}\n {{- '\\n{\"name\": \"' }}\n {{- tool_call.name }}\n {{- '\", \"arguments\": ' }}\n {%- if tool_call.arguments is string %}\n {{- tool_call.arguments }}\n {%- else %}\n {{- tool_call.arguments | tojson }}\n {%- endif %}\n {{- '}\\n' }}\n {%- endfor %}\n {%- endif %}\n {{- '<|im_end|>\\n' }}\n {%- elif message.role == \"tool\" %}\n {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n {{- '<|im_start|>user' }}\n {%- endif %}\n {{- '\\n\\n' }}\n {{- content }}\n {{- '\\n' }}\n {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n {{- '<|im_end|>\\n' }}\n {%- endif %}\n {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n {{- '<|im_start|>assistant\\n' }}\n {%- if enable_thinking is defined and enable_thinking is false %}\n {{- '\\n\\n\\n\\n' }}\n {%- endif %}\n{%- endif %}"
}
\ No newline at end of file
diff --git a/training_args.bin b/training_args.bin
new file mode 100644
index 0000000000000000000000000000000000000000..133cb6661b0124bb9db0d7d208d748fd736ea689
--- /dev/null
+++ b/training_args.bin
@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6696e1fef8bde5e22de2f5ae18c284f2b2d35a345157db5151571518385fb38b
+size 5713