Metraipolis › archive › dc_2852da52c74c
FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving — arXiv:2611.07542
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.07542
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-06T14:41:39.181Z
- SHA-256 of the archived text
ccdfa3fe5523cd39c5dab3fba9504f5bb206ccbeef514cdbb6b747495abe28f3- Stable URL of this page
https://metraipolis.com/documento/dc_2852da52c74c
How to cite
FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving — arXiv:2611.07542 — arXiv (en), https://arxiv.org/abs/2611.07542 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2852da52c74c (Metraipolis archive dc_2852da52c74c, sha256 ccdfa3fe5523cd39c5dab3fba9504f5bb206ccbeef514cdbb6b747495abe28f3)
BibTeX
@misc{dc2852da52c74c,
title = {FlashOffloading: Low-Latency KV Cache Offloading for Long-Context LLM Serving — arXiv:2611.07542},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.07542},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2852da52c74c; sha256 ccdfa3fe5523cd39c5dab3fba9504f5bb206ccbeef514cdbb6b747495abe28f3},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.07542. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
In large language model (LLM) serving systems, the key-value (KV) cache grows linearly with sequence length and concurrency, so long-context inference easily hits the high-bandwidth memory (HBM) capacity wall and results in low concurrency. Offloading the KV cache to host dynamic random-access memory (DRAM) frees HBM space and raises concurrency, but the additional cache management and host-to-device (H2D) load overhead inflates inference latency, higher concurrency fails to translate into effective throughput. We present FlashOffloading, a host-DRAM-centric inference architecture that fully offloads the KV cache in both the prefill and decode phases. FlashOffloading builds rolling HBM cache for prefill, a hot cache for decode, a reverse sparse index with zero extra miss-lookup overhead, and a compute--load overlapped kernel, optimizing KV offloading latency across the entire pipeline and delivering throughput gains under a time-per-output-token (TPOT) service-level objective (SLO). In experiments deploying GLM-5.2 with multi-token prediction (MTP3) speculative decoding on an Ascend910C prefill--decode (PD)-disaggregated platform, FlashOffloading improves prefill throughput by up to 1.38\(\times\) and decode throughput by up to 1.91\(\times\) under a 30 ms TPOT constraint. FlashOffloading is integrated into an open-source serving framework: this https URL . Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2611.07542 — 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 2026-11-06T14:18:42.641Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).