Metraipolis › archive › dc_3a951db60386

MLIPS: Multi-Layer Indexer Pool Sharing — arXiv:2610.27438

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.27438
License (as archived)
CC0-1.0
Archived
2026-10-21T07:07:44.675Z
SHA-256 of the archived text
d6fefe7d56e1fc30a8294356c5606d54c28e77e2e9802ea51b0599f73249f049
Stable URL of this page
https://metraipolis.com/documento/dc_3a951db60386

How to cite

MLIPS: Multi-Layer Indexer Pool Sharing — arXiv:2610.27438 — arXiv (en), https://arxiv.org/abs/2610.27438 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_3a951db60386 (Metraipolis archive dc_3a951db60386, sha256 d6fefe7d56e1fc30a8294356c5606d54c28e77e2e9802ea51b0599f73249f049)
BibTeX
@misc{dc3a951db60386,
  title = {MLIPS: Multi-Layer Indexer Pool Sharing — arXiv:2610.27438},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.27438},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_3a951db60386; sha256 d6fefe7d56e1fc30a8294356c5606d54c28e77e2e9802ea51b0599f73249f049},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.27438. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sparse attention limits the keys used by each query, but finding those keys can still require a full-prefix scan at every layer. We introduce \textbf{MLIPS (Multi-Layer Indexer Pool Sharing)}, a training-free method that shares candidate search while retaining each consumer's own scoring and selection. A producer publishes a top-$M$ pool larger than the attention budget $K$; subsequent layers select their own top-$K$ from that pool. Measurements across three sparse-attention model families show substantially higher coverage from broad pools than overlap between final selections. We implement layerwise sharing, persistent decode reuse, and a coarse producer in vLLM for DeepSeek-V4-Flash. Across four long-context suites, three consumer layers per producer give $1.14$--$1.40\times$ speedups and $14.8$--$27.4\%$ lower GPU energy, with observed accuracy changes of $-1.49$ to $+0.29$ percentage points in single runs. More extensive sharing exposes task-dependent quality losses. Sampled prefill traces also reveal recurring selected KV positions across query groups and layers. We characterize this shared working-set structure and use it to motivate Top-$G$, an exploratory bounded pool construction for prefill. Its deployment, task accuracy, and serving benefits remain open.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Performance (cs.PF).
Pagina del lavoro: https://arxiv.org/abs/2610.27438 — 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 2026-10-21T07:05:43.558Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).