MUR

MUR: Momentum Uncertainty guided Reasoning for Large Language Models

Paper Link: https://huggingface.co/papers/2507.14958

Code Repo: https://github.com/yayayacc/MUR

🔥 News

• 🔥🔥🔥 We release the paper as well as the code for MUR ! ! !

📖 Results

MUR reduces computation by over 50% on average across three backbone models, while improving accuracy by 0.62–3.37%.

<p align="center"> <img src="Intro.png" alt="scaling" width="400"> </p>

🚀 Quick Start

To use MUR, we can try with the following command.

Firstly, create the environment and install the requirements. This implementation is accelerated and supported by vllm.

# env
conda create -n mur python==3.11.9
conda activate mur
pip install -r requirements.txt

Next, simply run different python files:

python [TTS setting]-[vanilla|mur].py

Finally, run eval files. To be specific, please eval gpqa_diamond dataset using eval/eval_gpqa_cot.py. Adiitionaly, use eval/math_verifier.py to verify math datasets.

Feel free to contact with me if you have any questions ~~~

Citation

If you find it helpful, please kindly cite the paper.

@article{yan2025mur,
  title={MUR: Momentum Uncertainty guided Reasoning for Large Language Models},
  author={Hang Yan, Fangzhi Xu, Rongman Xu, Yifei Li, Jian Zhang, Haoran Luo, Xiaobao Wu, Luu Anh Tuan, Haiteng Zhao, Qika Lin, Jun Liu},
  journal={arXiv preprint arXiv:2507.14958},
  year={2025}
}