Metraipolis › archive › dc_dbb53ebd5a20

CEDAR: Error-Bounded Residual Routing for Efficient Long-Context Attention — arXiv:2609.07237

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.07237
License (as archived)
CC0-1.0
Archived
2026-10-16T12:45:30.499Z
SHA-256 of the archived text
07ed9a427a7bfa806032a6482b3673293e2527987fa5d9f1ceb829ed9f64c7b5
Stable URL of this page
https://metraipolis.com/documento/dc_dbb53ebd5a20

How to cite

CEDAR: Error-Bounded Residual Routing for Efficient Long-Context Attention — arXiv:2609.07237 — arXiv (en), https://arxiv.org/abs/2609.07237 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_dbb53ebd5a20 (Metraipolis archive dc_dbb53ebd5a20, sha256 07ed9a427a7bfa806032a6482b3673293e2527987fa5d9f1ceb829ed9f64c7b5)
BibTeX
@misc{dcdbb53ebd5a20,
  title = {CEDAR: Error-Bounded Residual Routing for Efficient Long-Context Attention — arXiv:2609.07237},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.07237},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_dbb53ebd5a20; sha256 07ed9a427a7bfa806032a6482b3673293e2527987fa5d9f1ceb829ed9f64c7b5},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.07237. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Post-hoc sparse attention accelerates long-context prefill by routing each query to a small set of token-level interactions. Hard selection, however, assigns zero probability to every omitted chunk: a routing miss cannot be recovered, and a fixed expansion budget spends the same work on easy and ambiguous queries. We introduce Coarse-to-fine Error-aware Dynamic Attention Routing (CEDAR), a coarse-to-fine method that keeps the language model frozen while preserving global coverage. Each semantic chunk contributes a cheap key--value summary to a residual attention path; chunks with high estimated approximation error are then expanded to exact token attention. Exact and summarized contributions are combined in a single softmax normalization, so refinement replaces, rather than duplicates, coarse evidence. We derive an output-error bound governed by within-chunk key/value dispersion and use it to allocate a variable refinement budget. A controlled clustered-attention study shows that residual summaries reduce reconstruction error by more than 98% relative to hard dropping at equal exact-chunk budgets. Experiments on long-context benchmarks demonstrate that CEDAR recovers most of the quality lost by hard sparse routing while maintaining approximately $3\times$ kernel speedup at 128K context.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-09 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2609.07237 — 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 2026-10-16T12:30:25.103Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-09?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).