Small Vision-Language Models are Smart Compressors for Long Video Understanding Paper β’ 2604.08120 β’ Published about 1 month ago β’ 20