{
  "id": 9055387,
  "title": "ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs",
  "url": "https://urgent.news/2026/09/20/valuediff-value-geometric-kv-cache-eviction-for-sink-suppressed-llms",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-20T03:03:33.000Z",
  "source": {
    "name": "arXiv cs.AI",
    "slug": "arxiv-cs-ai",
    "url": "https://arxiv.org/abs/2609.23314v1"
  },
  "original_language": "en",
  "account": null,
  "summary": "Modern LLMs with QK-normalization, gated attention, learned attention sinks, or logit softcapping exhibit weaker persistent attention sinks, on which existing KV cache eviction methods primarily rely. We observe that across these models, weaker sinks co-occur with greater value-vector dispersion relative to key-vector dispersion. Motivated by this value-side dispersion, we present ValueDiff, a…",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 1,
    "also_reported_by": []
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}