https://arxiv.org/abs/2510.08525 Which Heads Matter for Reasoning? RL-Guided KV Cache CompressionReasoning large language models exhibit complex reasoning behaviors via extended chain-of-thought generation that are highly fragile to information loss during decoding, creating critical challenges for KV cache compression. Existing token-dropping methodsarxiv.org ICLR은 떨어지고, ICML 2026에 붙은 것 같네요 Rea..
