KV-Efficient VLA: A Method of Speed up Vision Language Model with RNN-Gated Chunked KV Cache figure
AlphaXiv 中文概览(可滚动查看)