Metraipolis › archive › dc_cb2cecc3e134

AgentKV: Phase-Aware KV Eviction for Agentic LLMs — arXiv:2609.14872

Source: arXiv (en)
Original: https://arxiv.org/abs/2609.14872
License (as archived): CC0-1.0
Archived: 2026-10-15T23:17:08.957Z
SHA-256 of the archived text: f1f426ad806ad0150903dafcf900f27a112cfbaa8652f1da07750da12412a548
Stable URL of this page: https://metraipolis.com/documento/dc_cb2cecc3e134

How to cite

AgentKV: Phase-Aware KV Eviction for Agentic LLMs — arXiv:2609.14872 — arXiv (en), https://arxiv.org/abs/2609.14872 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_cb2cecc3e134 (Metraipolis archive dc_cb2cecc3e134, sha256 f1f426ad806ad0150903dafcf900f27a112cfbaa8652f1da07750da12412a548)

BibTeX

@misc{dccb2cecc3e134,
  title = {AgentKV: Phase-Aware KV Eviction for Agentic LLMs — arXiv:2609.14872},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.14872},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_cb2cecc3e134; sha256 f1f426ad806ad0150903dafcf900f27a112cfbaa8652f1da07750da12412a548},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.14872. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Agentic serving can consume orders of magnitude more tokens than chatbot workloads, stressing both KV-cache capacity and decode-time bandwidth. Most KV-eviction methods score cached keys against representative queries drawn from the most recent tokens, assuming future attention resembles recent attention. We show that agentic generation violates this assumption: future queries form a mixture over think, act, tool, and others phases, and principal-angle analysis shows these components occupy measurably different query subspaces, so recency representatives systematically undervalue keys that upcoming phases will need. We propose AGENTKV, which maintains a small query buffer per phase and scores cached keys against their union. We further implement AGENTKV in a persistent multi-turn serving path that carries compressed KV state across turns and compacts retained KV pages online. Across two models, six task domains, and three KV budgets each, AGENTKV improves task score by 5.5 points on average over R-KV and 5.3 over Tri-attention. Relative to upstream full-KV SGLang, AGENTKV improves output-token throughput by up to 1.80x. Code: this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-15 in cs.CL).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.14872 — 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 2026-10-15T22:51:10.029Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-15?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).