Model save

README.md

@@ -0,0 +1,67 @@
+---
+library_name: transformers
+model_name: Qwen2.5-3B-EMPO-TQA
+tags:
+- generated_from_trainer
+- trl
+- grpo
+licence: license
+---
+
+# Model Card for Qwen2.5-3B-EMPO-TQA
+
+This model is a fine-tuned version of [None](https://huggingface.co/None).
+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="qingyangzhang/Qwen2.5-3B-EMPO-TQA", 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/zqyoung1127-tianjin-university/huggingface/runs/gaqlrb6w) 
+
+
+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: 0.14.0
+- Transformers: 4.48.3
+- Pytorch: 2.5.1
+- Datasets: 3.1.0
+- 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": 0.00023655204281105568,
+    "train_runtime": 1010.415,
+    "train_samples": 490,
+    "train_samples_per_second": 0.485,
+    "train_steps_per_second": 0.01
+}

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": "4.48.3"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 0.0,
+    "train_loss": 0.00023655204281105568,
+    "train_runtime": 1010.415,
+    "train_samples": 490,
+    "train_samples_per_second": 0.485,
+    "train_steps_per_second": 0.01
+}

trainer_state.json

@@ -0,0 +1,162 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.975609756097561,
+  "eval_steps": 100,
+  "global_step": 10,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "completion_length": 19.635416507720947,
+      "epoch": 0.0975609756097561,
+      "grad_norm": 0.5056069493293762,
+      "kl": 0.0,
+      "learning_rate": 2e-06,
+      "loss": 0.0,
+      "reward": 0.5598958358168602,
+      "reward_std": 0.09434993285685778,
+      "rewards/semantic_entropy": 0.5598958358168602,
+      "step": 1
+    },
+    {
+      "completion_length": 17.5711807012558,
+      "epoch": 0.1951219512195122,
+      "grad_norm": 1.0753142833709717,
+      "kl": 0.0,
+      "learning_rate": 1.9396926207859082e-06,
+      "loss": 0.0,
+      "reward": 0.5069444291293621,
+      "reward_std": 0.130544594489038,
+      "rewards/semantic_entropy": 0.5069444291293621,
+      "step": 2
+    },
+    {
+      "completion_length": 17.8680557012558,
+      "epoch": 0.2926829268292683,
+      "grad_norm": 1.0308548212051392,
+      "kl": 0.0010528564453125,
+      "learning_rate": 1.766044443118978e-06,
+      "loss": 0.0,
+      "reward": 0.5251736156642437,
+      "reward_std": 0.13977272436022758,
+      "rewards/semantic_entropy": 0.5251736156642437,
+      "step": 3
+    },
+    {
+      "completion_length": 20.52256941795349,
+      "epoch": 0.3902439024390244,
+      "grad_norm": 0.6616184711456299,
+      "kl": 0.00194549560546875,
+      "learning_rate": 1.5e-06,
+      "loss": 0.0001,
+      "reward": 0.5691550932824612,
+      "reward_std": 0.13807372376322746,
+      "rewards/semantic_entropy": 0.5691550932824612,
+      "step": 4
+    },
+    {
+      "completion_length": 19.373263835906982,
+      "epoch": 0.4878048780487805,
+      "grad_norm": 0.7476760149002075,
+      "kl": 0.003894805908203125,
+      "learning_rate": 1.1736481776669305e-06,
+      "loss": 0.0002,
+      "reward": 0.5060763992369175,
+      "reward_std": 0.10967991594225168,
+      "rewards/semantic_entropy": 0.5060763992369175,
+      "step": 5
+    },
+    {
+      "completion_length": 19.572916984558105,
+      "epoch": 0.5853658536585366,
+      "grad_norm": 0.5244656205177307,
+      "kl": 0.0069637298583984375,
+      "learning_rate": 8.263518223330696e-07,
+      "loss": 0.0003,
+      "reward": 0.6232638955116272,
+      "reward_std": 0.11568193091079593,
+      "rewards/semantic_entropy": 0.6232638955116272,
+      "step": 6
+    },
+    {
+      "completion_length": 18.697916507720947,
+      "epoch": 0.6829268292682927,
+      "grad_norm": 0.5207864046096802,
+      "kl": 0.013153076171875,
+      "learning_rate": 5.000000000000002e-07,
+      "loss": 0.0005,
+      "reward": 0.621527798473835,
+      "reward_std": 0.1189112700521946,
+      "rewards/semantic_entropy": 0.621527798473835,
+      "step": 7
+    },
+    {
+      "completion_length": 19.454861402511597,
+      "epoch": 0.7804878048780488,
+      "grad_norm": 0.6792996525764465,
+      "kl": 0.0097503662109375,
+      "learning_rate": 2.339555568810221e-07,
+      "loss": 0.0004,
+      "reward": 0.5682870484888554,
+      "reward_std": 0.13645811565220356,
+      "rewards/semantic_entropy": 0.5682870484888554,
+      "step": 8
+    },
+    {
+      "completion_length": 20.125000476837158,
+      "epoch": 0.8780487804878049,
+      "grad_norm": 0.5297054052352905,
+      "kl": 0.010364532470703125,
+      "learning_rate": 6.030737921409168e-08,
+      "loss": 0.0004,
+      "reward": 0.5454282537102699,
+      "reward_std": 0.10149804083630443,
+      "rewards/semantic_entropy": 0.5454282537102699,
+      "step": 9
+    },
+    {
+      "completion_length": 19.699653148651123,
+      "epoch": 0.975609756097561,
+      "grad_norm": 0.4804926812648773,
+      "kl": 0.01203155517578125,
+      "learning_rate": 0.0,
+      "loss": 0.0005,
+      "reward": 0.6212384253740311,
+      "reward_std": 0.09729464584961534,
+      "rewards/semantic_entropy": 0.6212384253740311,
+      "step": 10
+    },
+    {
+      "epoch": 0.975609756097561,
+      "step": 10,
+      "total_flos": 0.0,
+      "train_loss": 0.00023655204281105568,
+      "train_runtime": 1010.415,
+      "train_samples_per_second": 0.485,
+      "train_steps_per_second": 0.01
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 10,
+  "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": 1,
+  "trial_name": null,
+  "trial_params": null
+}