Metraipolis › archive › dc_e1ffadb4a7cd

Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation — arXiv:2610.07716

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.07716
License (as archived)
CC0-1.0
Archived
2026-10-07T11:23:10.499Z
SHA-256 of the archived text
8d04d369477a108f57fae67ebc1da91b5036d49063b865492905a9880910f7be
Stable URL of this page
https://metraipolis.com/documento/dc_e1ffadb4a7cd

How to cite

Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation — arXiv:2610.07716 — arXiv (en), https://arxiv.org/abs/2610.07716 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_e1ffadb4a7cd (Metraipolis archive dc_e1ffadb4a7cd, sha256 8d04d369477a108f57fae67ebc1da91b5036d49063b865492905a9880910f7be)
BibTeX
@misc{dce1ffadb4a7cd,
  title = {Readout Stability in Prefill-Only Decision Models:Zero-Label Prediction and Inference-Time Compute Allocation — arXiv:2610.07716},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.07716},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_e1ffadb4a7cd; sha256 8d04d369477a108f57fae67ebc1da91b5036d49063b865492905a9880910f7be},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.07716. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Prefill-only decision models inspired by the Jev model score every candidate in a menu during a single forward pass and never decode, which makes one call one to two orders of magnitude cheaper than a same-scale generative language model. We show that this read-out structure comes with a testable property. When an intervention changes only the candidate menu and leaves the input text fixed, the post-intervention accuracy is already determined by the cached first-pass distribution. The estimator restricts the pass-1 probabilities to the menu, renormalizes, and reads off the argmax; it uses no labels and no second forward pass. Across seven model families, ten datasets and two task types, menu-only interventions are predicted to within 4.2 points, and for one family the prediction is exact. A probability-level variant of the same estimator errs by 21.0 points, so the property lives in the ranking rather than in the probabilities and is not recovered by calibration. Same-scale generative language models do not share the property. On those models the same estimator errs by 1.6 to 15.8 points and degrades as the model grows. The property turns inference-time compute into a decision that can be made before deployment. Uniform extra passes buy calibration but almost no accuracy; at matched cost a confidence cascade outperforms every scheme that re-asks the same model, and curating the menu beats enlarging the model, with a 0.8B model on a curated 5-candidate menu reaching 95.4% on CLINC150 against 80.0% for a 4B model on the full 150-label this http URL and data are 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/2610.07716 — 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 2026-10-07T11:15:39.427Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).