Metraipolis › archive › dc_78c29484da17
Emergent Structure in the Marginal Attention Space of Language Models — arXiv:2610.03109
- Source
- arXiv
- Original
- https://arxiv.org/abs/2610.03109
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-05T22:51:43.076Z
- SHA-256 of the archived text
b477fd22b60bac17e1c87916679974fded7fd929753e0b8eee72b9413c66cb82- Stable URL of this page
https://metraipolis.com/documento/dc_78c29484da17
How to cite
Emergent Structure in the Marginal Attention Space of Language Models — arXiv:2610.03109 — arXiv, https://arxiv.org/abs/2610.03109 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_78c29484da17 (Metraipolis archive dc_78c29484da17, sha256 b477fd22b60bac17e1c87916679974fded7fd929753e0b8eee72b9413c66cb82)
BibTeX
@misc{dc78c29484da17,
title = {Emergent Structure in the Marginal Attention Space of Language Models — arXiv:2610.03109},
howpublished = {arXiv},
url = {https://arxiv.org/abs/2610.03109},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_78c29484da17; sha256 b477fd22b60bac17e1c87916679974fded7fd929753e0b8eee72b9413c66cb82},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv, https://arxiv.org/abs/2610.03109. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
While representation similarity across independently trained language models is well-documented, how internal mechanics such as attention behave across models remains far less characterized. Inspired by this gap, we examine the structure of post-softmax attention weights by marginalizing over query positions, mapping them into a joint token-head "marginal attention space". Evaluating across 60+ diverse LLMs, we find that different properties emerge when reducing this space along its token and head axes. When reduced token-wise, marginal attention yields a text-intrinsic signal robustly conserved across models. To explain this property, we empirically connect marginal attention to the input-output Jacobian of the network, and prove theoretically that under a smoothness assumption, models with similar next-token distributions are guaranteed to have similar input-output Jacobian statistics. When reduced head-wise, it forms a model-private signature conserved across documents. Practically, this provides a natural way to estimate a per-head budget for key-value (KV) cache eviction, effectively decoupling model-specific budget allocation from text-intrinsic token scoring. On standard eviction benchmarks, a per-head budget precomputed offline on pretraining text, combined with a training-free token score, shows competitive performance with methods that recompute the budget on every document or train it per target. Code available at this https URL Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno). Categorie arXiv: Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.03109 — 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 2026-10-05T22:46:10.572Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).