Metraipolis › archive › dc_ad14858b9537

INTACT: Inducing Latent Reasoning on a Frozen Backbone — arXiv:2610.19568

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.19568
License (as archived)
CC0-1.0
Archived
2026-10-15T14:23:26.962Z
SHA-256 of the archived text
0e3a0a4fb13835d914f116da114826d69236e5c0bf47cf3a494db99167e8d11d
Stable URL of this page
https://metraipolis.com/documento/dc_ad14858b9537

How to cite

INTACT: Inducing Latent Reasoning on a Frozen Backbone — arXiv:2610.19568 — arXiv (en), https://arxiv.org/abs/2610.19568 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ad14858b9537 (Metraipolis archive dc_ad14858b9537, sha256 0e3a0a4fb13835d914f116da114826d69236e5c0bf47cf3a494db99167e8d11d)
BibTeX
@misc{dcad14858b9537,
  title = {INTACT: Inducing Latent Reasoning on a Frozen Backbone — arXiv:2610.19568},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19568},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ad14858b9537; sha256 0e3a0a4fb13835d914f116da114826d69236e5c0bf47cf3a494db99167e8d11d},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19568. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Latent reasoning replaces natural-language chains of thought with compact continuous "soft-token" representations, reducing the number of intermediate tokens that must be decoded autoregressively. We use a small external encoder to induce latent reasoning in a frozen backbone model. The encoder iteratively maps the backbone's hidden states to new latent blocks, constructing a latent reasoning prefix, after which the frozen model writes its remaining reasoning and its answer as text. Our method, INTACT, trains this encoder through distillation to reproduce the effect of written reasoning steps while keeping the model's behavior close to that of the frozen backbone, and then reinforces it on answer correctness. Across four frozen backbones from 1B to 8B, INTACT improves both GSM8K and average out-of-domain accuracy, including GSM8K gains from 75.8% to 79.3% on Llama-3.2-3B and from 77.6% to 79.7% on Qwen2.5-3B. It also reduces end-to-end latency per problem by 19% on Llama-3.2-3B and outperforms the evaluated latent-reasoning baselines at the 3B and 8B scales. Beyond the mathematics setting, the same recipe trained on a multi-domain corpus improves the frozen backbone on twelve evaluation benchmarks spanning mathematics, science, and commonsense, without task-specific adaptation.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.19568 — 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 2026-10-15T13:53:35.872Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).