Model save

README.md

@@ -0,0 +1,68 @@
+---
+base_model: aidando73/Qwen-2.5-7B-Simple-RL-v2
+library_name: transformers
+model_name: Qwen-2.5-7B-Simple-RL-v3
+tags:
+- generated_from_trainer
+- trl
+- grpo
+licence: license
+---
+
+# Model Card for Qwen-2.5-7B-Simple-RL-v3
+
+This model is a fine-tuned version of [aidando73/Qwen-2.5-7B-Simple-RL-v2](https://huggingface.co/aidando73/Qwen-2.5-7B-Simple-RL-v2).
+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="aidando73/Qwen-2.5-7B-Simple-RL-v3", 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/aidando73-personal/huggingface/runs/dlr1639u) 
+
+
+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": 0.1568015617286337,
+    "train_runtime": 9197.0383,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.815,
+    "train_steps_per_second": 0.006
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 2048,
+  "transformers_version": "4.49.0"
+}

train_results.json

@@ -0,0 +1,8 @@
+{
+    "total_flos": 0.0,
+    "train_loss": 0.1568015617286337,
+    "train_runtime": 9197.0383,
+    "train_samples": 7500,
+    "train_samples_per_second": 0.815,
+    "train_steps_per_second": 0.006
+}

trainer_state.json

@@ -0,0 +1,217 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 0.9893390191897654,
+  "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": 490.39622497558594,
+      "epoch": 0.017057569296375266,
+      "grad_norm": 0.48167672753334045,
+      "kl": 0.0,
+      "learning_rate": 5e-07,
+      "loss": 0.0927,
+      "reward": 1.4520089775323868,
+      "reward_std": 0.4626273363828659,
+      "rewards/accuracy_reward": 0.731026828289032,
+      "rewards/format_reward": 0.7209821715950966,
+      "step": 1
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 495.0767583847046,
+      "epoch": 0.08528784648187633,
+      "grad_norm": 0.48367658257484436,
+      "kl": 0.00019475817680358887,
+      "learning_rate": 2.5e-06,
+      "loss": 0.0758,
+      "reward": 1.3839286305010319,
+      "reward_std": 0.47902860026806593,
+      "rewards/accuracy_reward": 0.6654576193541288,
+      "rewards/format_reward": 0.7184710167348385,
+      "step": 5
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 469.24533462524414,
+      "epoch": 0.17057569296375266,
+      "grad_norm": 1055.7178955078125,
+      "kl": 0.09486055374145508,
+      "learning_rate": 2.956412726139078e-06,
+      "loss": 0.0696,
+      "reward": 1.559375074505806,
+      "reward_std": 0.39700488820672036,
+      "rewards/accuracy_reward": 0.6868303894996644,
+      "rewards/format_reward": 0.8725446864962578,
+      "step": 10
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 447.36073303222656,
+      "epoch": 0.255863539445629,
+      "grad_norm": 0.46586015820503235,
+      "kl": 0.008279037475585938,
+      "learning_rate": 2.7836719084521715e-06,
+      "loss": 0.068,
+      "reward": 1.6738840073347092,
+      "reward_std": 0.30874924659729003,
+      "rewards/accuracy_reward": 0.7267857491970062,
+      "rewards/format_reward": 0.9470982506871224,
+      "step": 15
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 444.303369140625,
+      "epoch": 0.3411513859275053,
+      "grad_norm": 0.7344509363174438,
+      "kl": 0.01377105712890625,
+      "learning_rate": 2.4946839873611927e-06,
+      "loss": 0.0591,
+      "reward": 1.6803572088479997,
+      "reward_std": 0.31110375486314296,
+      "rewards/accuracy_reward": 0.7267857447266579,
+      "rewards/format_reward": 0.9535714685916901,
+      "step": 20
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 480.18819122314454,
+      "epoch": 0.42643923240938164,
+      "grad_norm": 0.34691816568374634,
+      "kl": 0.0154571533203125,
+      "learning_rate": 2.1156192081791355e-06,
+      "loss": 0.0959,
+      "reward": 1.6316965013742446,
+      "reward_std": 0.3553690694272518,
+      "rewards/accuracy_reward": 0.6892857432365418,
+      "rewards/format_reward": 0.9424107536673546,
+      "step": 25
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 519.4330619812011,
+      "epoch": 0.511727078891258,
+      "grad_norm": 1.267974853515625,
+      "kl": 0.0434295654296875,
+      "learning_rate": 1.6808050203829845e-06,
+      "loss": 0.1268,
+      "reward": 1.561160784959793,
+      "reward_std": 0.3939851924777031,
+      "rewards/accuracy_reward": 0.6450893133878708,
+      "rewards/format_reward": 0.9160714745521545,
+      "step": 30
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 571.6105178833008,
+      "epoch": 0.5970149253731343,
+      "grad_norm": 2.5649266242980957,
+      "kl": 0.14027099609375,
+      "learning_rate": 1.2296174432791415e-06,
+      "loss": 0.1967,
+      "reward": 1.4618304252624512,
+      "reward_std": 0.5041121542453766,
+      "rewards/accuracy_reward": 0.5946428827941418,
+      "rewards/format_reward": 0.8671875417232513,
+      "step": 35
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 581.8879707336425,
+      "epoch": 0.6823027718550106,
+      "grad_norm": 10.305343627929688,
+      "kl": 0.2900634765625,
+      "learning_rate": 8.029152419343472e-07,
+      "loss": 0.234,
+      "reward": 1.3854911401867867,
+      "reward_std": 0.5583220794796944,
+      "rewards/accuracy_reward": 0.5649553827941418,
+      "rewards/format_reward": 0.8205357566475868,
+      "step": 40
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 593.7596252441406,
+      "epoch": 0.767590618336887,
+      "grad_norm": 12.518046379089355,
+      "kl": 0.358349609375,
+      "learning_rate": 4.3933982822017883e-07,
+      "loss": 0.2316,
+      "reward": 1.353571480512619,
+      "reward_std": 0.5827566109597683,
+      "rewards/accuracy_reward": 0.5622768118977547,
+      "rewards/format_reward": 0.791294677555561,
+      "step": 45
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 578.9877510070801,
+      "epoch": 0.8528784648187633,
+      "grad_norm": 13.292229652404785,
+      "kl": 0.45947265625,
+      "learning_rate": 1.718159615201853e-07,
+      "loss": 0.2462,
+      "reward": 1.3236607640981675,
+      "reward_std": 0.6153911039233207,
+      "rewards/accuracy_reward": 0.5506696633994579,
+      "rewards/format_reward": 0.7729911044239998,
+      "step": 50
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 556.6399826049804,
+      "epoch": 0.9381663113006397,
+      "grad_norm": 13.38757038116455,
+      "kl": 0.5525390625,
+      "learning_rate": 2.4570139579284723e-08,
+      "loss": 0.2526,
+      "reward": 1.3421875596046449,
+      "reward_std": 0.6172608032822609,
+      "rewards/accuracy_reward": 0.5716518081724644,
+      "rewards/format_reward": 0.7705357447266579,
+      "step": 55
+    },
+    {
+      "clip_ratio": 0.0,
+      "completion_length": 546.5475145975748,
+      "epoch": 0.9893390191897654,
+      "kl": 0.59619140625,
+      "reward": 1.3314732909202576,
+      "reward_std": 0.5954927864174048,
+      "rewards/accuracy_reward": 0.5625000235935053,
+      "rewards/format_reward": 0.7689732511838278,
+      "step": 58,
+      "total_flos": 0.0,
+      "train_loss": 0.1568015617286337,
+      "train_runtime": 9197.0383,
+      "train_samples_per_second": 0.815,
+      "train_steps_per_second": 0.006
+    }
+  ],
+  "logging_steps": 5,
+  "max_steps": 58,
+  "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": 16,
+  "trial_name": null,
+  "trial_params": null
+}