Metraipolis › archive › dc_af58f07cb2f9

Rationale Distribution Distillation via Flow Matching for Distributional Continuous Latent Reasoning — arXiv:2610.19871

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.19871
License (as archived)
CC0-1.0
Archived
2026-10-15T20:02:49.407Z
SHA-256 of the archived text
b09c6d5337bd1eac42a92ad2a1bece83826d51cd8675f00d13d2e7eb99193cc5
Stable URL of this page
https://metraipolis.com/documento/dc_af58f07cb2f9

How to cite

Rationale Distribution Distillation via Flow Matching for Distributional Continuous Latent Reasoning — arXiv:2610.19871 — arXiv (en), https://arxiv.org/abs/2610.19871 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_af58f07cb2f9 (Metraipolis archive dc_af58f07cb2f9, sha256 b09c6d5337bd1eac42a92ad2a1bece83826d51cd8675f00d13d2e7eb99193cc5)
BibTeX
@misc{dcaf58f07cb2f9,
  title = {Rationale Distribution Distillation via Flow Matching for Distributional Continuous Latent Reasoning — arXiv:2610.19871},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.19871},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_af58f07cb2f9; sha256 b09c6d5337bd1eac42a92ad2a1bece83826d51cd8675f00d13d2e7eb99193cc5},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.19871. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Sampling multiple chain-of-thought rationales from a language model can uncover correct solutions missed by a single attempt. However, these rationales are typically lengthy, making repeated autoregressive generation costly. Continuous latent reasoning offers a compact alternative, but point-valued methods condition repeated answer sampling on a single rationale, which can limit exploration of diverse paths and answer coverage. We thus propose rationale distribution distillation (RDD), training a conditional flow over fixed-length latent sequences by distilling from multiple textual reasoning paths. Furthermore, we expand the distillation targets through within-query mixup, interpolating between complete latent rationales for the same query. The flow nonautoregressively generates the entire latent sequence, and a frozen language model autoregressively decodes answers from sampled rationale contexts. Empirically, the resulting distributional reasoner achieves higher answer coverage than the evaluated point-valued latent reasoning baselines, particularly on advanced mathematical tasks.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.19871 — 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 2026-10-15T19:35:48.379Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).