arXiv cs.AI / cs.LG / cs.CL·13h agoOne Block, Multiple Depths: Recurrent Vision Transformers with Depth-Programmed Experts#vision-transformer#mixture-of-experts#dinov2AI research