Cache-to-Cache: Direct Semantic Communication Between Large Language Models arxiv.org 5 points by jonbaer 5 hours ago