热搜 · 最高第 2 · 05:31

Cache-to-Cache: Direct Semantic Communication Between LLMs (2025)