Metraipolis › archive › dc_0ebe28e548b1

When Does Adaptive Halting Win? Selection Rules Matter in Looped Language Models — arXiv:2611.02333

Source
arXiv (en)
Original
https://arxiv.org/abs/2611.02333
License (as archived)
CC0-1.0
Archived
2026-11-03T11:11:51.546Z
SHA-256 of the archived text
4bedc32971aca10fcea4e1633e395abbf939b79b4a7f6be9fb2f46f77f0c4ef4
Stable URL of this page
https://metraipolis.com/documento/dc_0ebe28e548b1

How to cite

When Does Adaptive Halting Win? Selection Rules Matter in Looped Language Models — arXiv:2611.02333 — arXiv (en), https://arxiv.org/abs/2611.02333 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_0ebe28e548b1 (Metraipolis archive dc_0ebe28e548b1, sha256 4bedc32971aca10fcea4e1633e395abbf939b79b4a7f6be9fb2f46f77f0c4ef4)
BibTeX
@misc{dc0ebe28e548b1,
  title = {When Does Adaptive Halting Win? Selection Rules Matter in Looped Language Models — arXiv:2611.02333},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2611.02333},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_0ebe28e548b1; sha256 4bedc32971aca10fcea4e1633e395abbf939b79b4a7f6be9fb2f46f77f0c4ef4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.02333. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Looped language models can stop iterating early, and recent work shows that post-hoc readouts of the loop trajectory often rival the learned exit gate of the looped model Ouro. Such comparisons reduce each method to one operating point chosen by a selection rule. We ask whether that rule decides the outcome. On a frozen Ouro-1.4B with four loops, we compared a trained stability probe ($\pi_1$-x) with Ouro's native gate, adapted to a per-question stopping decision, in pre-registered studies; the confirmatory study was designed after an exploratory finding on development data and evaluated once on a sealed test set of 8,517 questions. In the pre-registered confirmatory comparison, which selected operating points with a bootstrap-screened 1 pp accuracy-loss rule, the probe saved more of the four-loop budget than the native gate (+0.053 gross macro saving over three benchmarks, 95% CI [+0.049, +0.057]), mostly from CommonsenseQA, where the probe's test upper bound on accuracy loss exceeded 1 pp. Under a point-only rule, a pre-specified secondary analysis on the same test set reversed the preference (-0.022 [-0.026, -0.018]), with operating points that exceeded the budget. In pre-specified secondary comparisons, the probe showed no demonstrated advantage over an untrained KL divergence between successive loops (+0.00125 [-0.0041, +0.0065]); we emphasise this result post hoc and claim no equivalence. The probe also broke its budget on one of two out-of-distribution benchmarks (3.26 pp loss on Social IQa). Savings are estimated from stored full-depth traces, not measured inference time, and our screen carries no formal risk-control guarantee. We conclude that halting methods should be compared as signals together with selection procedures, with strong untrained baselines and out-of-distribution checks.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2611.02333 — 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 2026-11-03T10:59:52.413Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).