Training-Free Multimodal Large Language Model Orchestration

Tianyu Xie, Yuexiao Ma, Yuhang Wu +5

Building interactive omni-modal assistants often relies on end-to-end multimodal alignment to fuse heterogeneous modalities, which incurs substantial data and compute costs and lim…