Metraipolis › archive › dc_a55e5dc36844
Exact Memory-Time Optimization for Prefix-Cached Language Model Serving — arXiv:2610.02766
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.02766
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-05T23:09:50.863Z
- SHA-256 of the archived text
40f22247d2412fc84efd26c03326fcb2384bd23991a1074cf57499a71c411f88- Stable URL of this page
https://metraipolis.com/documento/dc_a55e5dc36844
How to cite
Exact Memory-Time Optimization for Prefix-Cached Language Model Serving — arXiv:2610.02766 — arXiv, https://arxiv.org/abs/2610.02766 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_a55e5dc36844 (Metraipolis archive dc_a55e5dc36844, sha256 40f22247d2412fc84efd26c03326fcb2384bd23991a1074cf57499a71c411f88)
BibTeX
@misc{dca55e5dc36844,
title = {Exact Memory-Time Optimization for Prefix-Cached Language Model Serving — arXiv:2610.02766},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.02766},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_a55e5dc36844; sha256 40f22247d2412fc84efd26c03326fcb2384bd23991a1074cf57499a71c411f88},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.02766. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Retaining language-model prefix states trades recomputation against storage time. Optimizing each cached block independently can overcount savings: a resident block is usable only when the required preceding prefix is also available. We introduce Prefix-Certificate Retention (PCR), an exact finite-trace formulation for static, grouped, reset-on-access timeouts. Usable-prefix rewards become nodes whose prerequisites are timeout thresholds and preceding hit certificates. The resulting maximum-weight closure reduces to one minimum cut, with graph size linear in the number of block lookups and timeout choices. A breakpoint theorem extends the construction to all nonnegative timeouts without discretization error. We also derive a linear-time-in-grid-size dynamic program for ordered timeouts and bounds that certify the cost of this restriction. Exhaustive small-instance checks and chronological replay of 39,632 public Mooncake requests validate the formulation. On the fixed grid, ordered timeouts attain the unrestricted training optimum in 118 of 120 trace-grouping-price cases. Heterogeneous retention improves several held-out memory-time tradeoffs, but finer training optimization does not uniformly improve transfer. The contribution is a tractable optimization model and an auditable benchmark for retention policies; the experiments measure usable prefix blocks and storage time, not GPU latency. 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/2610.02766 — 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 2026-10-05T23:01:38.630Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).