Metraipolis › archive › dc_8fbeda289ce7
ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs — arXiv:2609.23314
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.23314
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-09T07:07:08.657Z
- SHA-256 of the archived text
f40adc061f7101a21edfa5ac450455c33843cf7a3b0c20501c6617bc4d999019- Stable URL of this page
https://metraipolis.com/documento/dc_8fbeda289ce7
How to cite
ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs — arXiv:2609.23314 — arXiv (en), https://arxiv.org/abs/2609.23314 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_8fbeda289ce7 (Metraipolis archive dc_8fbeda289ce7, sha256 f40adc061f7101a21edfa5ac450455c33843cf7a3b0c20501c6617bc4d999019)
BibTeX
@misc{dc8fbeda289ce7,
title = {ValueDiff: Value-Geometric KV Cache Eviction for Sink-Suppressed LLMs — arXiv:2609.23314},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.23314},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_8fbeda289ce7; sha256 f40adc061f7101a21edfa5ac450455c33843cf7a3b0c20501c6617bc4d999019},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.23314. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
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 future attention. We evaluate under fixed cache budgets, with eviction at every block boundary during prefill and at every decoding step during generation. On RULER at a tight 2k token budget, ValueDiff retains 88-99% of dense across seven sink-suppressed models (best on 6 out of 7). On LongBench at the 4k budget, ValueDiff averages 92% retention across sink-suppressed models versus 83% for the strongest prior baseline. On MATH-500, ValueDiff is the strongest non-dense method on every sink-suppressed model tested at the 25% cache budget, outperforming prior methods by up to ~20 points on gated-attention models. Across all three benchmarks, value geometry emerges as the more reliable query-invariant eviction signal for sink-suppressed models. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.LG). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.23314 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-10-09T06:58:09.287Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).