Model save

README.md

@@ -0,0 +1,68 @@
+---
+base_model: Qwen/Qwen2.5-3B-Instruct
+library_name: transformers
+model_name: Qwen2.5-3B-Instruct-grpo-MATHDATA-E1
+tags:
+- generated_from_trainer
+- trl
+- grpo
+licence: license
+---
+
+# Model Card for Qwen2.5-3B-Instruct-grpo-MATHDATA-E1
+
+This model is a fine-tuned version of [Qwen/Qwen2.5-3B-Instruct](https://huggingface.co/Qwen/Qwen2.5-3B-Instruct).
+It has been trained using [TRL](https://github.com/huggingface/trl).
+
+## Quick start
+
+```python
+from transformers import pipeline
+
+question = "If you had a time machine, but could only go to the past or the future once and never return, which would you choose and why?"
+generator = pipeline("text-generation", model="chenggong1995/Qwen2.5-3B-Instruct-grpo-MATHDATA-E1", device="cuda")
+output = generator([{"role": "user", "content": question}], max_new_tokens=128, return_full_text=False)[0]
+print(output["generated_text"])
+```
+
+## Training procedure
+
+[<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="150" height="24"/>](https://wandb.ai/gongc1995-city-university-of-hong-kong/huggingface/runs/nut4sim8) 
+
+
+This model was trained with GRPO, a method introduced in [DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models](https://huggingface.co/papers/2402.03300).
+
+### Framework versions
+
+- TRL: 0.16.0.dev0
+- Transformers: 4.49.0
+- Pytorch: 2.5.1
+- Datasets: 3.3.2
+- Tokenizers: 0.21.0
+
+## Citations
+
+Cite GRPO as:
+
+```bibtex
+@article{zhihong2024deepseekmath,
+    title        = {{DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models}},
+    author       = {Zhihong Shao and Peiyi Wang and Qihao Zhu and Runxin Xu and Junxiao Song and Mingchuan Zhang and Y. K. Li and Y. Wu and Daya Guo},
+    year         = 2024,
+    eprint       = {arXiv:2402.03300},
+}
+
+```
+
+Cite TRL as:
+    
+```bibtex
+@misc{vonwerra2022trl,
+	title        = {{TRL: Transformer Reinforcement Learning}},
+	author       = {Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec},
+	year         = 2020,
+	journal      = {GitHub repository},
+	publisher    = {GitHub},
+	howpublished = {\url{https://github.com/huggingface/trl}}
+}
+```

all_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 0.0,
+    "train_loss": 0.023221049398088302,
+    "train_runtime": 8217.2755,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.913,
+    "train_steps_per_second": 0.009
+}

generation_config.json

@@ -0,0 +1,14 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "repetition_penalty": 1.05,
+  "temperature": 0.7,
+  "top_k": 20,
+  "top_p": 0.8,
+  "transformers_version": "4.49.0"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 0.0,
+    "train_loss": 0.023221049398088302,
+    "train_runtime": 8217.2755,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.913,
+    "train_steps_per_second": 0.009
+}

trainer_state.json

@@ -0,0 +1,275 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9984,
+  "eval_steps": 83,
+  "global_step": 78,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 686.6350830078125,
+      "epoch": 0.064,
+      "grad_norm": 0.493407666683197,
+      "kl": 0.00012958049774169922,
+      "learning_rate": 1.2499999999999999e-06,
+      "loss": -0.0056,
+      "reward": 1.0069444552063942,
+      "reward_std": 0.580621787160635,
+      "rewards/accuracy_reward": 0.3986111156642437,
+      "rewards/format_reward": 0.18090277863666415,
+      "rewards/tag_count_reward": 0.42743055820465087,
+      "step": 5
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 652.9118125915527,
+      "epoch": 0.128,
+      "grad_norm": 51.891143798828125,
+      "kl": 13.452931213378907,
+      "learning_rate": 1.995974293995239e-06,
+      "loss": 0.1344,
+      "reward": 1.3629340469837188,
+      "reward_std": 0.62072244733572,
+      "rewards/accuracy_reward": 0.3295138917863369,
+      "rewards/format_reward": 0.3850694507360458,
+      "rewards/tag_count_reward": 0.6483506999909878,
+      "step": 10
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 676.9017471313476,
+      "epoch": 0.192,
+      "grad_norm": 17.215152740478516,
+      "kl": 1.034814453125,
+      "learning_rate": 1.9510565162951534e-06,
+      "loss": 0.0138,
+      "reward": 1.5778646022081375,
+      "reward_std": 0.6299503356218338,
+      "rewards/accuracy_reward": 0.3361111145466566,
+      "rewards/format_reward": 0.4927083343267441,
+      "rewards/tag_count_reward": 0.7490451440215111,
+      "step": 15
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 641.2159820556641,
+      "epoch": 0.256,
+      "grad_norm": 0.4615473449230194,
+      "kl": 0.032623291015625,
+      "learning_rate": 1.858448793601866e-06,
+      "loss": 0.0025,
+      "reward": 1.623437511920929,
+      "reward_std": 0.6311033435165883,
+      "rewards/accuracy_reward": 0.3847222276031971,
+      "rewards/format_reward": 0.5031250067055225,
+      "rewards/tag_count_reward": 0.7355902805924416,
+      "step": 20
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 653.2670181274414,
+      "epoch": 0.32,
+      "grad_norm": 0.2847181260585785,
+      "kl": 0.04327239990234375,
+      "learning_rate": 1.7227948638273915e-06,
+      "loss": 0.0167,
+      "reward": 1.8901041924953461,
+      "reward_std": 0.6056328244507313,
+      "rewards/accuracy_reward": 0.395138892903924,
+      "rewards/format_reward": 0.6468750037252903,
+      "rewards/tag_count_reward": 0.8480902835726738,
+      "step": 25
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 616.6788215637207,
+      "epoch": 0.384,
+      "grad_norm": 0.27531540393829346,
+      "kl": 5.2246551513671875,
+      "learning_rate": 1.5508969814521024e-06,
+      "loss": 0.0613,
+      "reward": 2.1147569686174394,
+      "reward_std": 0.540030462294817,
+      "rewards/accuracy_reward": 0.37777778282761576,
+      "rewards/format_reward": 0.8107638955116272,
+      "rewards/tag_count_reward": 0.9262152895331383,
+      "step": 30
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 638.4989616394043,
+      "epoch": 0.448,
+      "grad_norm": 0.2678464651107788,
+      "kl": 0.0264862060546875,
+      "learning_rate": 1.3513748240813427e-06,
+      "loss": 0.0124,
+      "reward": 2.2046007037162783,
+      "reward_std": 0.47457179576158526,
+      "rewards/accuracy_reward": 0.3527777839452028,
+      "rewards/format_reward": 0.8920138970017433,
+      "rewards/tag_count_reward": 0.9598090335726738,
+      "step": 35
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 643.7434097290039,
+      "epoch": 0.512,
+      "grad_norm": 0.22826902568340302,
+      "kl": 0.0377349853515625,
+      "learning_rate": 1.1342332658176555e-06,
+      "loss": 0.0127,
+      "reward": 2.335156279802322,
+      "reward_std": 0.4647583156824112,
+      "rewards/accuracy_reward": 0.45034722685813905,
+      "rewards/format_reward": 0.9152777850627899,
+      "rewards/tag_count_reward": 0.9695312604308128,
+      "step": 40
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 629.6038261413574,
+      "epoch": 0.576,
+      "grad_norm": 0.32088664174079895,
+      "kl": 0.04580535888671875,
+      "learning_rate": 9.103606910965665e-07,
+      "loss": 0.008,
+      "reward": 2.414062523841858,
+      "reward_std": 0.40865729935467243,
+      "rewards/accuracy_reward": 0.5135416679084301,
+      "rewards/format_reward": 0.9260416761040687,
+      "rewards/tag_count_reward": 0.974479179084301,
+      "step": 45
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 656.8579933166504,
+      "epoch": 0.64,
+      "grad_norm": 2.3553261756896973,
+      "kl": 0.08855743408203125,
+      "learning_rate": 6.909830056250526e-07,
+      "loss": 0.0135,
+      "reward": 2.4554687619209288,
+      "reward_std": 0.38128191530704497,
+      "rewards/accuracy_reward": 0.5336805626749992,
+      "rewards/format_reward": 0.9430555611848831,
+      "rewards/tag_count_reward": 0.9787326440215111,
+      "step": 50
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 652.9923721313477,
+      "epoch": 0.704,
+      "grad_norm": 2.0868406295776367,
+      "kl": 0.0515777587890625,
+      "learning_rate": 4.871007225940939e-07,
+      "loss": 0.015,
+      "reward": 2.487239581346512,
+      "reward_std": 0.3608484253287315,
+      "rewards/accuracy_reward": 0.5642361134290695,
+      "rewards/format_reward": 0.9427083373069763,
+      "rewards/tag_count_reward": 0.9802951425313949,
+      "step": 55
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 650.6635467529297,
+      "epoch": 0.768,
+      "grad_norm": 0.19568555057048798,
+      "kl": 0.035595703125,
+      "learning_rate": 3.0893735101313535e-07,
+      "loss": 0.0199,
+      "reward": 2.501822918653488,
+      "reward_std": 0.35273106172680857,
+      "rewards/accuracy_reward": 0.5711805634200573,
+      "rewards/format_reward": 0.947569452226162,
+      "rewards/tag_count_reward": 0.9830729216337204,
+      "step": 60
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 637.9875061035157,
+      "epoch": 0.832,
+      "grad_norm": 0.23244014382362366,
+      "kl": 0.03612060546875,
+      "learning_rate": 1.6542674627869734e-07,
+      "loss": 0.0137,
+      "reward": 2.5010416865348817,
+      "reward_std": 0.34778057783842087,
+      "rewards/accuracy_reward": 0.5725694492459297,
+      "rewards/format_reward": 0.9472222268581391,
+      "rewards/tag_count_reward": 0.9812500059604645,
+      "step": 65
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 633.2107666015625,
+      "epoch": 0.896,
+      "grad_norm": 71.81304168701172,
+      "kl": 0.2681488037109375,
+      "learning_rate": 6.376512936026279e-08,
+      "loss": 0.0135,
+      "reward": 2.534635454416275,
+      "reward_std": 0.3476743776351213,
+      "rewards/accuracy_reward": 0.5899305626749992,
+      "rewards/format_reward": 0.9600694462656975,
+      "rewards/tag_count_reward": 0.9846354201436043,
+      "step": 70
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 632.5708358764648,
+      "epoch": 0.96,
+      "grad_norm": 0.19250567257404327,
+      "kl": 0.038299560546875,
+      "learning_rate": 9.050238232065299e-09,
+      "loss": 0.0194,
+      "reward": 2.515277796983719,
+      "reward_std": 0.37559727430343626,
+      "rewards/accuracy_reward": 0.5836805649101734,
+      "rewards/format_reward": 0.9520833417773247,
+      "rewards/tag_count_reward": 0.9795139014720917,
+      "step": 75
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 630.8084576924642,
+      "epoch": 0.9984,
+      "kl": 0.040715535481770836,
+      "reward": 2.5170717736085257,
+      "reward_std": 0.3583106442044179,
+      "rewards/accuracy_reward": 0.5885416728754839,
+      "rewards/format_reward": 0.9496527860562006,
+      "rewards/tag_count_reward": 0.978877325852712,
+      "step": 78,
+      "total_flos": 0.0,
+      "train_loss": 0.023221049398088302,
+      "train_runtime": 8217.2755,
+      "train_samples_per_second": 0.913,
+      "train_steps_per_second": 0.009
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 78,
+  "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": 0.0,
+  "train_batch_size": 12,
+  "trial_name": null,
+  "trial_params": null
+}