ZipVL: Efficient Large Vision-Language Models with Dynamic Token Sparsification and KV Cache Compression Paper • 2410.08584 • Published Oct 11, 2024 • 12