AuroraZengfh/MM-MergeBench

Card of Dataset for Multimodal Large Language Model

Dataset details and sources

This dataset is constructed using publicly available instruction tuning datasets, including COCO, ScienceQA, VizWiz, ImageNet, VQAv2, ImageNet-R, Flickr30k, OCRVQA, Screen2words and TabMWP.

Seen datasets for merging

Dataset	Image Source	Download Path
ScienceQA	ScienceQA	images
VizWiz	VizWiz	images
ImageNet	ImageNet	images
VQAv2, Flickr30k	COCO2014	images
IconQA	IconQA	images
Flickr30k	Flickr30k	images
OCRVQA	OCRVQA	images

Unseen datasets for merging

Dataset	Image Source	Download Path
AOKVQA	COCO2014	images
ImageNet-R	ImageNet-R	images
Screen2words	Screen2words	images
TabMWP	TabMWP	images

Please download these datasets before proceeding with instruction tuning.

---

RobustMerge

Paper: RobustMerge: Parameter-Efficient Model Merging for MLLMs with Direction Robustness

Code: github.com/AuroraZengfh/RobustMerge

If you find this work useful, consider citing our paper as follows:

@article{zeng2025parameter,
  title={RobustMerge: Parameter-Efficient Model Merging for MLLMs with Direction Robustness},
  author={Zeng, Fanhu and Guo, Haiyang and Zhu, Fei and Shen, Li and Tang, Hao},
  journal={arXiv preprint arXiv:2502.17159},
  year={2025}
}

---

license: mit