Metraipolis › archive › dc_4ffa42d09f8c

SGD-KV: Summarization Guided KV Cache Compression — arXiv:2609.03235

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.03235
License (as archived)
CC0-1.0
Archived
2026-10-17T01:40:14.260Z
SHA-256 of the archived text
4b3ea43190072d53aaec2ed579ba5c9d475fa7fd7b8720fba423aff852180e19
Stable URL of this page
https://metraipolis.com/documento/dc_4ffa42d09f8c

How to cite

SGD-KV: Summarization Guided KV Cache Compression — arXiv:2609.03235 — arXiv (en), https://arxiv.org/abs/2609.03235 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_4ffa42d09f8c (Metraipolis archive dc_4ffa42d09f8c, sha256 4b3ea43190072d53aaec2ed579ba5c9d475fa7fd7b8720fba423aff852180e19)
BibTeX
@misc{dc4ffa42d09f8c,
  title = {SGD-KV: Summarization Guided KV Cache Compression — arXiv:2609.03235},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.03235},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_4ffa42d09f8c; sha256 4b3ea43190072d53aaec2ed579ba5c9d475fa7fd7b8720fba423aff852180e19},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.03235. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large language models (LLMs) face severe memory bottlenecks in long-context inference due to the linearly growing size of key-value (KV) caches. Existing KV cache compression techniques typically rely on simple heuristics, overlooking the distinct functional roles of different attention heads. We present SGD-KV (Summarization-Guided KV Cache Compression), a head-aware framework that leverages a novel chunk-summarization diagnostic task to systematically identify and prioritize attention heads specialized in hierarchical information aggregation. Experiments on Qwen2.5-7B-1M and Qwen3-32B across diverse long-context benchmarks demonstrate that SGD-KV achieves state-of-the-art performance with contexts up to 1M tokens, while reducing KV cache memory usage by up to 75%. Our findings show that strategically allocating the KV cache budget based on the summarization score distribution of attention heads yields a superior efficiency-accuracy trade-off for long-context inference.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-04 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.03235 — 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 2026-10-17T01:11:43.510Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-04?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).