Fetching the paper…

SmartTrim: Adaptive Tokens and Attention Pruning for Efficient Vision-Language Models · Around