3 ms·Cache-to-Cache: Direct Semantic Communication Between Large Language Models14 points by jonbaer 11mo ago