Last released Jun 5, 2026
Training-free visual token sparsification for vision-language models (ICML 2025)
Supported by