Robust Multimodal Large Language Models Against Modality Conflict

Official repository for
Robust Multimodal Large Language Models Against Modality Conflict

---

🌟 Overview

This repository provides the code and dataset for our paper:
Robust Multimodal Large Language Models Against Modality Conflict.

• GitHub Repository: zmzhang2000/MMMC
• Hugging Face Hub: ustc-zhangzm/MMMC
• Paper: OpenReview PDF

---

📦 Multimodal Modality Conflict (MMMC) Dataset

The MMMC dataset is available on the Hugging Face Hub. You can easily download and use it as follows:

from datasets import load_dataset

dataset = load_dataset("ustc-zhangzm/MMMC")

Note: The dataset is generated by large language models and may contain some noise. We recommend using the dataset for research purposes only.

---

🚀 Improving the Robustness of MLLMs

We provide code for supervised fine-tuning and reinforcement learning to enhance the robustness of Multimodal Large Language Models (MLLMs) under modality conflict scenarios.

• Please follow the documentation for instructions on running the code.
• Detailed explanations of these methods are available in our paper.

---

📄 License

This dataset is distributed under the CC BY-SA 3.0 license.

---

📖 Citation

If you find this work helpful for your research, please cite our paper:

@inproceedings{
    zhang2025robust,
    title={Robust Multimodal Large Language Models Against Modality Conflict},
    author={Zongmeng Zhang and Wengang Zhou and Jie Zhao and Houqiang Li},
    booktitle={Forty-second International Conference on Machine Learning},
    year={2025},
    url={https://openreview.net/forum?id=SP43jVv7fJ}
}

---