Prompt Optimization with Human Feedback

This is the official implementation of the paper Prompt Optimization with Human Feedback.

Oral Presentation at ICML 2024 Workshop on Models of Human Feedback for AI Alignment

Video

This repo is based on the codebase of INSTINCT.

Prepare the data

Please download the data from: https://github.com/xqlin98/INSTINCT/tree/main/Induction/experiments/data/instruction_induction and put it under ./experiments/data/instruction_induction

Prepare the environment

We use conda to manage our environment. Please install our environment using the following command: conda env create -f environment.yml

Prepare your OpenAI API key

Add your OpenAI key to Induction/key

Find our running scripts

Our running scripts are in experiments/run_dbandits_po.sh, experiments/run_dbandits_image_gen.sh and experiments/run_dbandits_response.sh. To run the script

cd Induction
bash experiments/run_dbandits_po.sh

Citation

If you find this repo/paper helpful, please consider citing our paper:

@article{lin2024prompt,
  title={Prompt Optimization with Human Feedback},
  author={Lin, Xiaoqiang and Dai, Zhongxiang and Verma, Arun and Ng, See-Kiong and Jaillet, Patrick and Low, Bryan Kian Hsiang},
  journal={arXiv preprint arXiv:2405.17346},
  year={2024}
}