Better Zero-Shot Reasoning with Role-Play Prompting

This is the code for our paper "Better Zero-Shot Reasoning with Role-Play Prompting". Data is from the repo of Zero-Shot-CoT

The repository is the latest version. After the paper is officially published, we will update its arxiv version.

Environment

openai 0.27.8
torch 2.0.1
pandas 2.0.3

Usage

First, you need an OpenAI API key and write it into utils.py.

# Sentence Generator (Decoder) for GPT-3 ...
def decoder_for_gpt3(args, input):
    
    # GPT-3 API allows each users execute the API within 60 times in a minute ...
    time.sleep(1)

    # enter your OpenAI key below
    openai.api_key = ""

Then you can run the code using run.sh.

bash run.sh

More details are presented in run.sh. And We provide our experimental results in my_log folder.

We observe that the results sometimes may not be completely fixed though the temperature is set to 0. But this does not affect our conclusion. See related discussions in OpenAI.

citation

@misc{kong2023better,
      title={Better Zero-Shot Reasoning with Role-Play Prompting}, 
      author={Aobo Kong and Shiwan Zhao and Hao Chen and Qicheng Li and Yong Qin and Ruiqi Sun and Xin Zhou},
      year={2023},
      eprint={2308.07702},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}