#moe-vie

MoE-ViE targets efficient image and video understanding

The paper studies mixture-of-experts scaling for CLIP-style vision encoders.