luka-group / mDPO

[EMNLP 2024] mDPO: Conditional Preference Optimization for Multimodal Large Language Models.
https://feiwang96.github.io/mDPO/
27 stars 0 forks source link

Multimodal Direct Preference Optimization (mDPO)

mDPO: Conditional Preference Optimization for Multimodal Large Language Models (published at EMNLP 2024).

🌐 Homepage | 📖 Paper | 💻 Code | 🤗 Dataset

alt text

Updates

Installation

TBD

Training

Our training data is available at this link.

To train Bunny with mDPO, use the following command:

python bunny/run_mdpo_bunny.py

Evaluation

TBD

Citation

Please cite the following paper if you find the repo helpful:

@article{wang2024mdpo,
  title={mDPO: Conditional Preference Optimization for Multimodal Large Language Models},
  author={Wang, Fei and Zhou, Wenxuan and Huang, James Y and Xu, Nan and Zhang, Sheng and Poon, Hoifung and Chen, Muhao},
  journal={arXiv preprint arXiv:2406.11839},
  year={2024}
}