Introduction

• Paper: MM-IFEngine: Towards Multimodal Instruction Following

• Github: SYuan03/MM-IFEngine

• Project Page: syuan03.github.io/MM-IFEngine/

• MM-IFEval Evaluation: Using VLMEvalKit

😊 This is the official repo of MM-IFEngine datasets in MM-IFEngine: Towards Multimodal Instruction Following

🚀 We include both the SFT and DPO data in this repo as the v1 dataset (generated mainly by InternVL2.5-78B and Qwen2-VL-7B), which we used to train the model described in our paper.

💖 [2025.9.16 Update] We have released the v2 dataset (annotated mainly by GPT-4o), feel free to use it!

Format

Using ShareGPT format from LLaMA-Factory

Citation

@article{ding2025mm,
  title={MM-IFEngine: Towards Multimodal Instruction Following},
  author={Ding, Shengyuan and Wu, Shenxi and Zhao, Xiangyu and Zang, Yuhang and Duan, Haodong and Dong, Xiaoyi and Zhang, Pan and Cao, Yuhang and Lin, Dahua and Wang, Jiaqi},
  journal={arXiv preprint arXiv:2504.07957},
  year={2025}
}