Metraipolis › archive › dc_214a4dceeb0a

Language Models Can Control Their Own Attention — arXiv:2609.02737

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.02737
License (as archived)
CC0-1.0
Archived
2026-10-17T03:38:32.742Z
SHA-256 of the archived text
0a2e73f05c2c67a66ac890b750235e91cd6e082e7b9af08b33096b672a739de9
Stable URL of this page
https://metraipolis.com/documento/dc_214a4dceeb0a

How to cite

Language Models Can Control Their Own Attention — arXiv:2609.02737 — arXiv (en), https://arxiv.org/abs/2609.02737 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_214a4dceeb0a (Metraipolis archive dc_214a4dceeb0a, sha256 0a2e73f05c2c67a66ac890b750235e91cd6e082e7b9af08b33096b672a739de9)
BibTeX
@misc{dc214a4dceeb0a,
  title = {Language Models Can Control Their Own Attention — arXiv:2609.02737},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.02737},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_214a4dceeb0a; sha256 0a2e73f05c2c67a66ac890b750235e91cd6e082e7b9af08b33096b672a739de9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.02737. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Language models spend most of their attention on a small fraction of context, yet they read the entire KV cache to find the few tokens that matter. If the user asks about a previous detail in a 1M-token conversation, global attention layers must scan the full context to generate each token of the reply. A prominent approach pre-selects relevant tokens via lightweight proxy scores, but this extrinsic scoring still incurs O(N) per step. We take an intrinsic approach motivated by the simple question: wouldn't the model already know which parts of the context are relevant? We introduce Declarative Attention (DA), a protocol that elicits the model to declare where it needs to attend within its chain-of-thought, partitioning generation into three modes: <global> (full context), <focus> (a specific region), and <local> (recent output only). The inference engine parses these declarations like tool calls and skips most of the KV cache read. Under zero-shot evaluation across 15 long-context tasks, DA on off-the-shelf models (Gemma-4-31B, Qwen-3.6-27B) reduces total attended tokens per response (45.8%, 19.6%), despite extra decoding steps, with modest accuracy drops (1.40pp, 2.10pp) that shrink with model scale. DA unlocks a new axis of sparse attention, with further potential under training-based methods. Code is available at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-03 in cs.AI).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.02737 — 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 2026-10-17T03:13:12.567Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-03?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).