Seeing the Forest and the Trees: Query-Aware Tokenizer for Long-Video Multimodal Language Models Paper โข 2511.11910 โข Published Nov 14 โข 35