Xuerui2312/DeepSeek-R1-Distill-Qwen-7B-TRPA-DeepScaleR-verl0326

📝 Trust Region Preference Approximation: A simple and stable reinforcement learning algorithm for LLM reasoning

Checkpoints of Trust Region Preference Approximation (TRPA) on DeepScaleR based on DeepSeek-R1-Distill-Qwen-7B using the 0326 version of verl0326

[TRPA Paper | TRPA Github]

📚 Citation

@article{su2025trust,
  title={Trust region preference approximation: A simple and stable reinforcement learning algorithm for llm reasoning},
  author={Su, Xuerui and Xie, Shufang and Liu, Guoqing and Xia, Yingce and Luo, Renqian and Jin, Peiran and Ma, Zhiming and Wang, Yue and Wang, Zun and Liu, Yuting},
  journal={arXiv preprint arXiv:2504.04524},
  year={2025}
}