Metraipolis › archive › dc_7eab75c75359

One Model, Two Paths: Balancing Single-Pass and Looped Performance in Language Models — arXiv:2610.33517

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.33517
License (as archived)
CC0-1.0
Archived
2026-10-27T08:11:22.028Z
SHA-256 of the archived text
e800259e04f6e72f57e0c46101f63da7780f72b72f02c3ae70d9a41a334affbb
Stable URL of this page
https://metraipolis.com/documento/dc_7eab75c75359

How to cite

One Model, Two Paths: Balancing Single-Pass and Looped Performance in Language Models — arXiv:2610.33517 — arXiv (en), https://arxiv.org/abs/2610.33517 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_7eab75c75359 (Metraipolis archive dc_7eab75c75359, sha256 e800259e04f6e72f57e0c46101f63da7780f72b72f02c3ae70d9a41a334affbb)
BibTeX
@misc{dc7eab75c75359,
  title = {One Model, Two Paths: Balancing Single-Pass and Looped Performance in Language Models — arXiv:2610.33517},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.33517},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_7eab75c75359; sha256 e800259e04f6e72f57e0c46101f63da7780f72b72f02c3ae70d9a41a334affbb},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.33517. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Repeating Transformer layers can improve language-model performance without adding parameters. A pretrained model can be adapted to support both single-pass and looped inference by reusing middle-layer weights for additional computation. Because both paths share weights, training to improve looped predictions can degrade single-pass quality. We study joint training strategies using weighted losses on both outputs, with guidance from the frozen original model. Joint supervision substantially reduces single-pass degradation compared to looped-only training, while preserving looped gains over ordinary continued training. Most of this benefit comes from adjusting the single-pass loss weight, with guidance from the frozen original model further improving the looped output. With validation-selected weights, our method preserves single-pass quality within a predefined tolerance of ordinary continued training while improving looped predictions.

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/2610.33517 — 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 2026-10-27T07:56:00.201Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).