Metraipolis › archive › dc_2fb49235e58e

AttentionIndex: Accelerating Learned Sparse Attention with a Causal Inverted-File Index — arXiv:2611.02508

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.02508
License (as archived)
CC0-1.0
Archived
2026-11-03T10:53:33.188Z
SHA-256 of the archived text
14c8ae59a14d68d1ee84dbc018ab4a88cda732ae7d21e95f0f8ead22a14c0189
Stable URL of this page
https://metraipolis.com/documento/dc_2fb49235e58e

How to cite

AttentionIndex: Accelerating Learned Sparse Attention with a Causal Inverted-File Index — arXiv:2611.02508 — arXiv (en), https://arxiv.org/abs/2611.02508 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2fb49235e58e (Metraipolis archive dc_2fb49235e58e, sha256 14c8ae59a14d68d1ee84dbc018ab4a88cda732ae7d21e95f0f8ead22a14c0189)
BibTeX
@misc{dc2fb49235e58e,
  title = {AttentionIndex: Accelerating Learned Sparse Attention with a Causal Inverted-File Index — arXiv:2611.02508},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02508},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2fb49235e58e; sha256 14c8ae59a14d68d1ee84dbc018ab4a88cda732ae7d21e95f0f8ead22a14c0189},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02508. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Learned sparse attention, used in production models such as DeepSeek-V4 and GLM-5, lets each query attend to only k keys chosen by a lightweight indexer, but the indexer still scores every key in the prefix. At one million tokens, this selection takes more than half of prefill time. Two-stage selection over contiguous blocks, as in HiSA, reduces this cost, but its recall declines as context grows at a fixed candidate budget. We present AttentionIndex, which routes each query over an inverted-file index that groups indexer keys by similarity and refines the candidates with the native indexer score. AttentionIndex builds the index during prefill, making new keys searchable immediately, reusing previous clustering to keep maintenance cheap, and exposing each query only to clusters built from its own prefix. On GLM-5.3 and DeepSeek-V4-Flash at one million tokens, AttentionIndex delivers 1.3-1.8 times the prefill throughput of exhaustive selection, and at throughput within 5% of HiSA, it recovers 17.0 and 10.0 percentage points more of the exhaustive top-k, respectively. AttentionIndex also improves accuracy on long-context benchmarks by up to 18.4 percentage points over HiSA at the same candidate budget.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2611.02508 — 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 2026-11-03T10:49:38.354Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).