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_05-new
+tags:
+- generated_from_trainer
+- trl
+- grpo
+licence: license
+---
+
+# Model Card for DeepSeek-R1-Distill-Qwen-7B-RL-length-penalty-low-medium-0_05-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_05-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/xdifirou) 
+
+
+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": 0.1985041233229226,
+    "train_runtime": 18468.5879,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.812,
+    "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": 0.1985041233229226,
+    "train_runtime": 18468.5879,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.812,
+    "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": 0.26080044292213134,
+      "kl": 0.0,
+      "learning_rate": 5e-07,
+      "loss": 0.1215,
+      "reward": 0.7184314727783203,
+      "reward_std": 0.15251445909962058,
+      "rewards/accuracy_reward": 0.7421875298023224,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.02375606936402619,
+      "step": 1
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 975.6433238983154,
+      "epoch": 0.1702127659574468,
+      "grad_norm": 0.2057173239852816,
+      "kl": 7.790327072143555e-05,
+      "learning_rate": 2.5e-06,
+      "loss": 0.1143,
+      "reward": 0.7099975757300854,
+      "reward_std": 0.147755601326935,
+      "rewards/accuracy_reward": 0.7338169943541288,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.02381941577186808,
+      "step": 5
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 866.1272705078125,
+      "epoch": 0.3404255319148936,
+      "grad_norm": 0.4006290505834594,
+      "kl": 0.005752372741699219,
+      "learning_rate": 2.956412726139078e-06,
+      "loss": 0.1039,
+      "reward": 0.7401266694068909,
+      "reward_std": 0.16346940137445926,
+      "rewards/accuracy_reward": 0.7612723588943482,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.02114568497054279,
+      "step": 10
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 727.8928909301758,
+      "epoch": 0.5106382978723404,
+      "grad_norm": 0.789066138623355,
+      "kl": 0.174072265625,
+      "learning_rate": 2.7836719084521715e-06,
+      "loss": 0.1145,
+      "reward": 0.6513140380382538,
+      "reward_std": 0.1966270312666893,
+      "rewards/accuracy_reward": 0.6690848499536515,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.017770822765305637,
+      "step": 15
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 573.2581741333008,
+      "epoch": 0.6808510638297872,
+      "grad_norm": 1.576084537592535,
+      "kl": 0.8208984375,
+      "learning_rate": 2.4946839873611927e-06,
+      "loss": 0.1556,
+      "reward": 0.585111603885889,
+      "reward_std": 0.22033293917775154,
+      "rewards/accuracy_reward": 0.5991071686148643,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.013995561096817256,
+      "step": 20
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 543.6825035095214,
+      "epoch": 0.851063829787234,
+      "grad_norm": 2.632876303519198,
+      "kl": 2.15634765625,
+      "learning_rate": 2.1156192081791355e-06,
+      "loss": 0.3664,
+      "reward": 0.606927415728569,
+      "reward_std": 0.2152379907667637,
+      "rewards/accuracy_reward": 0.6202009156346321,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.013273498858325183,
+      "step": 25
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 515.0230895996094,
+      "epoch": 1.0340425531914894,
+      "grad_norm": 1.7973794229669702,
+      "kl": 1.01513671875,
+      "learning_rate": 1.6808050203829845e-06,
+      "loss": 0.1836,
+      "reward": 0.6228495962917805,
+      "reward_std": 0.21852513104677201,
+      "rewards/accuracy_reward": 0.6353794872760773,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.01252990250941366,
+      "step": 30
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 475.1066070556641,
+      "epoch": 1.2042553191489362,
+      "grad_norm": 3.0368202315739117,
+      "kl": 1.555859375,
+      "learning_rate": 1.2296174432791415e-06,
+      "loss": 0.2713,
+      "reward": 0.559271278232336,
+      "reward_std": 0.24244545102119447,
+      "rewards/accuracy_reward": 0.5708705581724643,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011599282221868634,
+      "step": 35
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 469.2449981689453,
+      "epoch": 1.374468085106383,
+      "grad_norm": 3.5433702308800723,
+      "kl": 1.345703125,
+      "learning_rate": 8.029152419343472e-07,
+      "loss": 0.2314,
+      "reward": 0.5598608091473579,
+      "reward_std": 0.26654755510389805,
+      "rewards/accuracy_reward": 0.5713169902563096,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011456176661886275,
+      "step": 40
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 475.7722343444824,
+      "epoch": 1.5446808510638297,
+      "grad_norm": 1.1370291260510161,
+      "kl": 1.3955078125,
+      "learning_rate": 4.3933982822017883e-07,
+      "loss": 0.2259,
+      "reward": 0.5786746814846992,
+      "reward_std": 0.2622571483254433,
+      "rewards/accuracy_reward": 0.5902902036905289,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011615533195436002,
+      "step": 45
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 474.67189865112306,
+      "epoch": 1.7148936170212767,
+      "grad_norm": 1.7209049738299256,
+      "kl": 1.1173828125,
+      "learning_rate": 1.718159615201853e-07,
+      "loss": 0.1973,
+      "reward": 0.5861792124807834,
+      "reward_std": 0.25209669955074787,
+      "rewards/accuracy_reward": 0.5977678894996643,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011588669382035732,
+      "step": 50
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 478.20337142944334,
+      "epoch": 1.8851063829787233,
+      "grad_norm": 1.3684947080685528,
+      "kl": 1.382421875,
+      "learning_rate": 2.4570139579284723e-08,
+      "loss": 0.2311,
+      "reward": 0.5730349622666836,
+      "reward_std": 0.24952978417277336,
+      "rewards/accuracy_reward": 0.5847098492085934,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011674887221306562,
+      "step": 55
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 478.4012222290039,
+      "epoch": 1.9872340425531916,
+      "kl": 1.4332682291666667,
+      "reward": 0.5787627287209034,
+      "reward_std": 0.2581109864016374,
+      "rewards/accuracy_reward": 0.5904018133878708,
+      "rewards/format_reward": 0.0,
+      "rewards/len_penalty": -0.011639068795678517,
+      "step": 58,
+      "total_flos": 0.0,
+      "train_loss": 0.1985041233229226,
+      "train_runtime": 18468.5879,
+      "train_samples_per_second": 0.812,
+      "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
+}