Mitigating the Bias of Large Language Model Evaluation

This is the official repository for paper Mitigating the Bias of Large Language Model Evaluation.

In this paper, we propose systematic research about the bias of LLM-as-a-Judge. Specifically, for closed-source judge models, we apply calibration to mitigate the significance of superficial quality, both on probability level and prompt level. For open-source judge models, we propose to mitigate the bias by contrastive training, with curated negative samples that deviate from instruction but present better superficial quality.

Citation

@inproceedings{zhou2024mitigating,
    title={Mitigating the Bias of Large Language Model Evaluation},
    author={Zhou, Hongli and Huang, Hui and Long, Yunfei and Xu, Bing and Zhu, Conghui and Cao, Hailong and Yang, Muyun and Zhao, Tiejun},
    booktitle={The 23rd China National Conference on Computational Linguistics},
    year={2024}
}

Acknowledgement

This repo benefits from JudgeLM and LLMBar. Thanks for their wonderful works.