MisMitiFact Critique Models

This repository contains the trained critique model weights used in the paper:
Generating Grounded Responses to Counter Misinformation via Learning Efficient Fine-Grained Critiques.

Overview

Critique models provide fine-grained feedback on initial responses generated by large language models.
They are a key part of the MisMitiFact framework, guiding LLMs to produce more grounded and misinformation-resistant outputs.

This repo contains only the model weights for the trained critique models.
For inference code, preprocessing scripts, and experiment pipelines, please see the main MisMitiFact repository.

Citation

If you use these models in your work, please cite:

@misc{xu2025generatinggroundedresponsescounter,
      title={Generating Grounded Responses to Counter Misinformation via Learning Efficient Fine-Grained Critiques}, 
      author={Xiaofei Xu and Xiuzhen Zhang and Ke Deng},
      year={2025},
      eprint={2506.05924},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2506.05924}, 
}