Metraipolis › archive › dc_78ba3671adae

Block-Sparse Attention with Semantic-Geometric Decoupled Routing — arXiv:2609.22884

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.22884
License (as archived)
CC0-1.0
Archived
2026-10-09T06:49:39.746Z
SHA-256 of the archived text
2e25667a0a83fec3f67bee72110e27c5de7a296a34d5f1a1a6715e08dc3d43e6
Stable URL of this page
https://metraipolis.com/documento/dc_78ba3671adae

How to cite

Block-Sparse Attention with Semantic-Geometric Decoupled Routing — arXiv:2609.22884 — arXiv (en), https://arxiv.org/abs/2609.22884 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_78ba3671adae (Metraipolis archive dc_78ba3671adae, sha256 2e25667a0a83fec3f67bee72110e27c5de7a296a34d5f1a1a6715e08dc3d43e6)
BibTeX
@misc{dc78ba3671adae,
  title = {Block-Sparse Attention with Semantic-Geometric Decoupled Routing — arXiv:2609.22884},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.22884},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_78ba3671adae; sha256 2e25667a0a83fec3f67bee72110e27c5de7a296a34d5f1a1a6715e08dc3d43e6},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.22884. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Long-context inference has become a defining capability of large language models, but exact dense attention remains costly due to its quadratic scaling with sequence length. Block-sparse attention offers a hardware-friendly alternative by routing each query block to a small set of relevant key blocks, yet accurate training-free block routing remains difficult. Existing routers often pool post-RoPE token representations, which entangles semantic aggregation with RoPE-induced geometry and attenuates local positional cues through high-frequency phase cancellation. To resolve this mismatch, we propose \textbf{Semantic-Geometric Decoupled Routing}, a training-free block routing framework that shifts semantic aggregation to the pre-RoPE space and reconstructs geometric bias with an offline structural prior and relative block distances. This decomposition yields an explicit closed-form block routing score without token-level search or post-hoc calibration. Experiments on long-context text and video tasks show that our method approaches full-attention accuracy across 4K--128K contexts, keeps routing overhead below 3.4 ms, and achieves a 5.03$\times$ speedup over FlashAttn at a 128K context length.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-22 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2609.22884 — 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 2026-10-09T06:48:44.961Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-22?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).