Reproduce “OmniMoE-VL: A Sparse Vision-Language Model with Coupled Visual-Depth Routing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Long Qian, Bingke Zhu, Jiaqi Wei, Yingying Chen, Jinqiao Wang
OmniMoE-VL routes question-conditioned visual evidence from intermediate encoder depths to both language-model input fusion and updates within language layers. It selects a shared global route, refining that preference locally for fusion. In the authors’ eight-benchmark comparison, its average is 85.9 versus an aggregate-derived 80.3 baseline; matched controls favor shared routing, but do not establish faster inference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.