arXiv preprint / cs.CV

DecoMoE: Decoupling Visual Propagation and Expert Computation for Efficient Multimodal MoE Inference

作者：Xudong Tan, Peng Ye, Ming Xie, Chenyu Huang, Yaoxin Yang, Jiayuan Fan, Tao Chen · 单位：Fudan University, The Chinese University of Hong Kong · 会议/期刊：arXiv preprint · 方向：cs.CV · 发布日期：2026-10-01

2026-10-01视觉语言
新标签打开 AlphaXiv新标签打开 Code
DecoMoE: Decoupling Visual Propagation and Expert Computation for Efficient Multimodal MoE Inference figure
AlphaXiv 中文概览（可滚动查看）