SOURCE-LINKED INTELLIGENCE

ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs

arXiv · AI, language, vision and robotics · article · Sep 20, 2026 · UTC

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 value-geometric eviction that ranks tokens by the L2 deviation of their value vectors from the cache mean. The same score arises as the minimal-disturbance eviction under a max-entropy assumption about f

Read original source ↗ Open in workspace

recordType

paper

region

Global