See What You Are Told: Visual Attention Sink in Large Multimodal Models

VLM
Re-assign community