Cache-to-Cache: Direct Semantic Communication Between Large Language Models arxiv.org 12 points by jonbaer a day ago