Learning to Predict Middle-Layer Attention in MLLMs for Visual Token Prunin figure
AlphaXiv 中文概览(可滚动查看)