2025 ICML ICML 2025

SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference