TRAN

This repo contains the codes for our work “Failures Pave the Way: Enhancing Large Language Models through Tuning-free Rule Accumulation” (EMNLP 2023).

Setup

The required package can be installed by running the following command.

pip install -r requirements.txt

For all experiments, please download the data from their repos (Big-Bench and TweetEval). The downloaded data are placed in the data folder. Here is the example of downloading Big-Bench.

cd data
git clone https://github.com/google/BIG-bench

Experiments

Please modify the openai key in the main.py. The scripts folder contains code to reproduce our experiments. For example, to run experiments on BBQ-Lite, run the following code:

bash scripts/run_bbq.sh # run experiments on bbq-lite

Contacts

Thank you for your interest in our work!

Please feel free to ask about any questions about the algorithms, codes, as well as problems encountered in running them so that we can make it clearer and better. You can either create an issue in the github repo or contact us at YangZeyuan2020@gmail.com.

Citation

@misc{yang2023failures,
      title={Failures Pave the Way: Enhancing Large Language Models through Tuning-free Rule Accumulation}, 
      author={Zeyuan Yang and Peng Li and Yang Liu},
      year={2023},
      eprint={2310.15746},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Licence

TRAN is licensed under the terms of the MIT license. See LICENSE for more details.