Model save

README.md

@@ -0,0 +1,68 @@
+---
+base_model: zijianh/DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-new
+library_name: transformers
+model_name: DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-medium-0_01-new
+tags:
+- generated_from_trainer
+- trl
+- grpo
+licence: license
+---
+
+# Model Card for DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-medium-0_01-new
+
+This model is a fine-tuned version of [zijianh/DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-new](https://huggingface.co/zijianh/DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-new).
+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="zijianh/DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-medium-0_01-new", 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/sota_mavens-university-of-michigan/huggingface/runs/cpkc5xj7) 
+
+
+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: 3.4.1
+- Tokenizers: 0.21.1
+
+## 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": 0.17393977451941062,
+    "train_runtime": 17719.3192,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.847,
+    "train_steps_per_second": 0.003
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 151646,
+  "do_sample": true,
+  "eos_token_id": 151643,
+  "temperature": 0.6,
+  "top_p": 0.95,
+  "transformers_version": "4.49.0"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 0.0,
+    "train_loss": 0.17393977451941062,
+    "train_runtime": 17719.3192,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.847,
+    "train_steps_per_second": 0.003
+}

trainer_state.json

@@ -0,0 +1,230 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.9872340425531916,
+  "eval_steps": 100,
+  "global_step": 58,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 973.0485916137695,
+      "epoch": 0.03404255319148936,
+      "grad_norm": 0.25066120729021635,
+      "kl": 0.0,
+      "learning_rate": 5e-07,
+      "loss": 0.1125,
+      "reward": 0.737436331808567,
+      "reward_std": 0.1497634956613183,
+      "rewards/accuracy_reward": 0.7421875298023224,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.004751214059069753,
+      "step": 1
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 979.3401679992676,
+      "epoch": 0.1702127659574468,
+      "grad_norm": 0.2302441340574829,
+      "kl": 7.614493370056152e-05,
+      "learning_rate": 2.5e-06,
+      "loss": 0.1016,
+      "reward": 0.7308486774563789,
+      "reward_std": 0.1540980595164001,
+      "rewards/accuracy_reward": 0.7356306128203869,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.004781934621860273,
+      "step": 5
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 873.104167175293,
+      "epoch": 0.3404255319148936,
+      "grad_norm": 0.42359185734828586,
+      "kl": 0.00587923526763916,
+      "learning_rate": 2.956412726139078e-06,
+      "loss": 0.1009,
+      "reward": 0.7584600493311882,
+      "reward_std": 0.1573723241686821,
+      "rewards/accuracy_reward": 0.762723246216774,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.004263204050948844,
+      "step": 10
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 745.0973556518554,
+      "epoch": 0.5106382978723404,
+      "grad_norm": 62.01063791631681,
+      "kl": 0.424346923828125,
+      "learning_rate": 2.7836719084521715e-06,
+      "loss": 0.1137,
+      "reward": 0.6722547233104705,
+      "reward_std": 0.1982373122125864,
+      "rewards/accuracy_reward": 0.6758928880095482,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.003638170822523534,
+      "step": 15
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 586.5627494812012,
+      "epoch": 0.6808510638297872,
+      "grad_norm": 0.9091305232027161,
+      "kl": 0.769775390625,
+      "learning_rate": 2.4946839873611927e-06,
+      "loss": 0.1393,
+      "reward": 0.601042202860117,
+      "reward_std": 0.20822684429585933,
+      "rewards/accuracy_reward": 0.603906275331974,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002864075929392129,
+      "step": 20
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 558.4970115661621,
+      "epoch": 0.851063829787234,
+      "grad_norm": 4.53697058337715,
+      "kl": 0.884765625,
+      "learning_rate": 2.1156192081791355e-06,
+      "loss": 0.1679,
+      "reward": 0.6045274563133717,
+      "reward_std": 0.21284419894218445,
+      "rewards/accuracy_reward": 0.6072544872760772,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002727036183932796,
+      "step": 25
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 530.7533882141113,
+      "epoch": 1.0340425531914894,
+      "grad_norm": 3.8562770589516298,
+      "kl": 0.818994140625,
+      "learning_rate": 1.6808050203829845e-06,
+      "loss": 0.1617,
+      "reward": 0.6154981851577759,
+      "reward_std": 0.22405075579881667,
+      "rewards/accuracy_reward": 0.6180803872644901,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002582195942522958,
+      "step": 30
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 497.5663185119629,
+      "epoch": 1.2042553191489362,
+      "grad_norm": 22.866090135660507,
+      "kl": 1.26787109375,
+      "learning_rate": 1.2296174432791415e-06,
+      "loss": 0.212,
+      "reward": 0.5886419266462326,
+      "reward_std": 0.24773433208465576,
+      "rewards/accuracy_reward": 0.5910714566707611,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.0024295230163261295,
+      "step": 35
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 494.5519203186035,
+      "epoch": 1.374468085106383,
+      "grad_norm": 2.82751872248447,
+      "kl": 1.95224609375,
+      "learning_rate": 8.029152419343472e-07,
+      "loss": 0.264,
+      "reward": 0.5949066542088985,
+      "reward_std": 0.25708167143166066,
+      "rewards/accuracy_reward": 0.5973214522004128,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002414804318686947,
+      "step": 40
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 510.7599586486816,
+      "epoch": 1.5446808510638297,
+      "grad_norm": 1.1188472228674733,
+      "kl": 1.22158203125,
+      "learning_rate": 4.3933982822017883e-07,
+      "loss": 0.2053,
+      "reward": 0.5883542984724045,
+      "reward_std": 0.25047608427703383,
+      "rewards/accuracy_reward": 0.5908482417464256,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002493945072637871,
+      "step": 45
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 516.5424369812011,
+      "epoch": 1.7148936170212767,
+      "grad_norm": 1.5793754040342751,
+      "kl": 1.18623046875,
+      "learning_rate": 1.718159615201853e-07,
+      "loss": 0.2085,
+      "reward": 0.6019421294331551,
+      "reward_std": 0.2471614670008421,
+      "rewards/accuracy_reward": 0.6044643111526966,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002522179845254868,
+      "step": 50
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 522.2845085144043,
+      "epoch": 1.8851063829787233,
+      "grad_norm": 1.202040205779595,
+      "kl": 1.11318359375,
+      "learning_rate": 2.4570139579284723e-08,
+      "loss": 0.2133,
+      "reward": 0.609391774237156,
+      "reward_std": 0.23280140571296215,
+      "rewards/accuracy_reward": 0.6119419939815998,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002550217299722135,
+      "step": 55
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 531.5718307495117,
+      "epoch": 1.9872340425531916,
+      "kl": 1.1866861979166667,
+      "reward": 0.620554693043232,
+      "reward_std": 0.23846940882503986,
+      "rewards/accuracy_reward": 0.6231399103999138,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.002585208606130133,
+      "step": 58,
+      "total_flos": 0.0,
+      "train_loss": 0.17393977451941062,
+      "train_runtime": 17719.3192,
+      "train_samples_per_second": 0.847,
+      "train_steps_per_second": 0.003
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 58,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 2,
+  "save_steps": 10,
+  "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": 32,
+  "trial_name": null,
+  "trial_params": null
+}