/CoT_benchmark

Code for "Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective"

Primary LanguagePythonMIT LicenseMIT

Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective

This repository is the official implementation of the paper “Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective”, based on the implementation of minGPT in PyTorch.

@misc{feng2023revealing,
      title={Towards Revealing the Mystery behind Chain of Thought: a Theoretical Perspective}, 
      author={Guhao Feng and Bohang Zhang and Yuntian Gu and Haotian Ye and Di He and Liwei Wang},
      year={2023},
      eprint={2305.15408},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Data Generation

The training and testing data can be generated via the command in DATA.md.

Training

The training instruction is in TRAIN.md.

Testing

The testing instruction is in TEST.md.

Training Logs

To faciliate future research, we upload all training logs in the logs folder.

License

This project is under the MIT license. See LICENSE for details.