Metraipolis › archive › dc_715bbca7ee24

OVAL: Output-Aware Local Page Bases for KV Cache Retrieval — arXiv:2610.06686

Source: arXiv
Original: https://arxiv.org/abs/2610.06686
License (as archived): CC0-1.0
Archived: 2026-10-06T05:31:45.349Z
SHA-256 of the archived text: 8550754b79a0f726a9bb58db4c5d62be566aa52c758fcbf659b50626b0463ef8
Stable URL of this page: https://metraipolis.com/documento/dc_715bbca7ee24

How to cite

OVAL: Output-Aware Local Page Bases for KV Cache Retrieval — arXiv:2610.06686 — arXiv, https://arxiv.org/abs/2610.06686 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_715bbca7ee24 (Metraipolis archive dc_715bbca7ee24, sha256 8550754b79a0f726a9bb58db4c5d62be566aa52c758fcbf659b50626b0463ef8)

BibTeX

@misc{dc715bbca7ee24,
  title = {OVAL: Output-Aware Local Page Bases for KV Cache Retrieval — arXiv:2610.06686},
  howpublished = {arXiv},
  url = {https://arxiv.org/abs/2610.06686},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_715bbca7ee24; sha256 8550754b79a0f726a9bb58db4c5d62be566aa52c758fcbf659b50626b0463ef8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.06686. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long context inference with large language models becomes increasingly expensive as attention must operate over an ever growing KV cache. Page sparse attention reduces this cost by representing each KV page compactly and retrieving only a subset for each query. Existing retrieval methods are designed to estimate attention scores or page relevance, but their objectives do not directly account for how approximation errors affect the resulting value weighted attention output. We introduce \method{}, an output aware page encoding derived from the joint structure of keys and values while preserving the key information needed for accurate retrieval. \method{} is training free and requires no additional value dependent statistics at inference time. Once constructed, its stored representation has the same size and decode time scoring cost as a key only spectral representation. Across long reasoning, long context understanding, and long generation benchmarks, \method{} consistently improves over the key only spectral baseline and performs competitively with recent KV cache compression and retrieval methods. On long reasoning benchmarks, it achieves strong avg@\(k\) performance across model benchmark pairs, while matching or surpassing leading baselines on several long context understanding and generation settings with modest decoding overhead. Code is available at \url{ this https URL }.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.06686 — 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 2026-10-06T05:27:05.906Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).