CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM 文章

ArXiv CS.AI2026-05-26NEWSen作者: Yubo Li, Yidi Miao
查看原文 →
关系图谱

CONF-KV: Confidence-Aware KV Cache Eviction with Mixed-Precision Storage for Long-Horizon LLM · 相关技术

相关技术

LLMCONF-KVKV Cachepyramidal per-layer budget variantmixed FP16/INT8 storageblockwise online-softmax attention