Small Vision-Language Models are Smart Compressors for Long Video Understanding Paper β’ 2604.08120 β’ Published 3 days ago β’ 13