Metraipolis › archive › dc_431aebfc0f66

Think Less, Locally: Local Preference Learning for Reasoning Compression — arXiv:2611.04814

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.04814
License (as archived)
CC0-1.0
Archived
2026-11-04T18:34:47.558Z
SHA-256 of the archived text
2be557223f4d70365ec22a24b16f30b527070ace727bd369f576eec88a50b7ea
Stable URL of this page
https://metraipolis.com/documento/dc_431aebfc0f66

How to cite

Think Less, Locally: Local Preference Learning for Reasoning Compression — arXiv:2611.04814 — arXiv (en), https://arxiv.org/abs/2611.04814 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_431aebfc0f66 (Metraipolis archive dc_431aebfc0f66, sha256 2be557223f4d70365ec22a24b16f30b527070ace727bd369f576eec88a50b7ea)
BibTeX
@misc{dc431aebfc0f66,
  title = {Think Less, Locally: Local Preference Learning for Reasoning Compression — arXiv:2611.04814},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.04814},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_431aebfc0f66; sha256 2be557223f4d70365ec22a24b16f30b527070ace727bd369f576eec88a50b7ea},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04814. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Large reasoning models achieve strong performance on challenging tasks, but their long chain-of-thought often incurs substantial inference costs by continuing to deliberate after sufficient information has been obtained. Existing reasoning compression methods primarily operate on reasoning trajectories, paying less attention to the local decision of whether further reasoning is necessary given the current state. We propose LoCo, a local preference learning method for reasoning compression. LoCo detects candidate overthinking states from the model's own traces, freezes the corresponding reasoning prefixes, regenerates shorter verified continuations from the same states, and trains on the resulting state-conditioned preference pairs with DPO. This formulation teaches the model when to continue reasoning and when to conclude, rather than simply encouraging shorter trajectories overall. We construct training data entirely from self-generated mathematical traces and evaluate LoCo on five mathematical and three code generation benchmarks. LoCo reduces completion length by 41.7\%--45.4\% on mathematics and 23.4\%--36.3\% on code, while maintaining or improving accuracy and achieving the best AUC across all code benchmarks. LoCo consistently improves the accuracy-efficiency trade-off on mathematics and transfers effectively to code generation despite math-only training, demonstrating the cross-domain generality of local preference learning for reasoning compression. The code and model will be publicly released at this https URL .

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.04814 — 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 2026-11-04T18:32:37.339Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).