Video-XL-2: Towards Very Long-Video Understanding Through Task-Aware KV Sparsification

Open in new window