[EMNLP 2025 Findings] 
MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation
Authors: Shoubin Yu*, Yue Zhang*, Ziyang Wang, Jaehong Yoon, Mohit Bansal
- Jun 22, 2025. Check our arXiv-version for MEXA.
- We will release multi-expert skills/captions code / data later
conda create -n mexa python=3.10
conda activate mexa
pip install -r requirements.txtWe provide MEXA inference script examples as follows.
sh run_mexa.shPlease cite our paper if you use our models in your work:
@article{yu2025mexa,
title={MEXA: Towards General Multimodal Reasoning with Dynamic Multi-Expert Aggregation},
author={Yu, Shoubin and Zhang, Yue and Wang, Ziyang and Yoon, Jaehong and Bansal, Mohit},
journal={arxiv:2506.17113},
year={2025}
}