BeaconKV: Key-Value Cache Compression Guided by Beacon Queries for Efficient Large Reasoning Model Inference

  • 类型:arxiv
  • 标识:2609.04971
  • 链接:https://arxiv.org/abs/2609.04971
  • 主分类:llm-infra
  • 形态:method
  • 被引:0
  • 被引来源:Semantic Scholar + OpenAlex
  • S2被引:0
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:BeaconKV is proposed, a training-free KV cache compression method that maintains beacon queries, compact representatives for each global query cluster, to anticipate which KV pairs will be revisited without storing the entire query history.
  • OpenAlex ID:W7211878501
  • OpenAlex DOI:10.48550/arxiv.2609.04971
  • DOI:10.48550/arxiv.2609.04971
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://arxiv.org/pdf/2609.04971
  • OpenAlex更新:2026-09-10
  • 待LLM分类:否
  • 标题中文:BeaconKV:由 Beacon 查询引导的 KV Cache 压缩,用于高效大型推理模型推理
  • TLDR中文:本文提出 BeaconKV,一种免训练的 KV cache 压缩方法,通过为每个全局查询簇维护紧凑代表性 beacon query 来预测哪些 KV 对将被重访,无需存储完整查询历史。
  • 来源文件:
  • /inbox/tom/_candidates/2026-09-09-agent-rag-longcontext-candidates.json
  • [OpenAlex backfill]
  • [S2 enrich]