Description

Official implementation of the paper: Dyve: Thinking Fast and Slow for Dynamic Process Verification

Setup

To set up the environment, run the following commands:

conda create -n dyve python==3.10.13
conda activate dyve

pip install -r requirements.txt

Data

Our data: https://huggingface.co/datasets/Jianyuan1/cot-data

Model

Our model: https://huggingface.co/Jianyuan1/deepseek-r1-14b-cot-math-reasoning-full/tree/main

Citation

If you find our work useful, please consider citing:

@misc{zhong2025dyvethinkingfastslow,
      title={Dyve: Thinking Fast and Slow for Dynamic Process Verification}, 
      author={Jianyuan Zhong and Zeju Li and Zhijian Xu and Xiangyu Wen and Qiang Xu},
      year={2025},
      eprint={2502.11157},
      archivePrefix={arXiv},
      primaryClass={cs.AI},
      url={https://arxiv.org/abs/2502.11157}, 
}

Acknowledge

https://github.com/QwenLM/ProcessBench

https://github.com/openreasoner/openr

https://github.com/QwenLM/Qwen2.5-Math

https://github.com/ZubinGou/math-evaluation-harness