Metraipolis › archive › dc_49f200d3141b

Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It — arXiv:2609.36585

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.36585
License (as archived)
CC0-1.0
Archived
2026-10-07T16:22:46.399Z
SHA-256 of the archived text
e1da0c5c6a964919e7628d1fcdacbc21cb61cf66175b91f1d9baa195fe0822e9
Stable URL of this page
https://metraipolis.com/documento/dc_49f200d3141b

How to cite

Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It — arXiv:2609.36585 — arXiv (en), https://arxiv.org/abs/2609.36585 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_49f200d3141b (Metraipolis archive dc_49f200d3141b, sha256 e1da0c5c6a964919e7628d1fcdacbc21cb61cf66175b91f1d9baa195fe0822e9)
BibTeX
@misc{dc49f200d3141b,
  title = {Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It — arXiv:2609.36585},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.36585},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_49f200d3141b; sha256 e1da0c5c6a964919e7628d1fcdacbc21cb61cf66175b91f1d9baa195fe0822e9},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.36585. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Pretrained transformers use little of their depth to follow references in context. Thirteen base models reliably follow only 1.4-3.6 lines, and extra pretrained loops add little. A task-trained rank-8 LoRA at one early layer extends this computation with all model weights frozen. Qwen3-8B improves from 15.5% to 99% exact accuracy on 24-line chains; a longer-trained LoRA reaches 50 lines. Ouro-1.4B reaches 60 lines after four loops and at least 160 after eight. The LoRA starts a relay: program lines pass on their chain identity through a short range of middle layers. Frozen heads read progressively further up the chain, and removing parent-line attention stops the relay. A frozen-model measurement locates the last useful intervention layer within tolerance in three of four held-out models. Task-specific LoRAs also improve MuSiQue. Default answers therefore understate the computation accessible through a tiny edit. Code and an interactive demo are available at this https URL

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-30 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL); Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.36585 — 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 2026-10-07T15:54:53.515Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-30?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).