Metraipolis › archive › dc_f668cc5ac03b

Sampling via Decision-Flow: Training-Free Extraction of Improved Latent Reasoning Paths in Large Language Models — arXiv:2609.12317

Source
arXiv (en)
Original
https://arxiv.org/abs/2609.12317
License (as archived)
CC0-1.0
Archived
2026-10-16T03:15:37.796Z
SHA-256 of the archived text
d45727496f86d1620ec90c60a19f84ba5ac52a8f1bb8f6104daf23d0e776e465
Stable URL of this page
https://metraipolis.com/documento/dc_f668cc5ac03b

How to cite

Sampling via Decision-Flow: Training-Free Extraction of Improved Latent Reasoning Paths in Large Language Models — arXiv:2609.12317 — arXiv (en), https://arxiv.org/abs/2609.12317 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f668cc5ac03b (Metraipolis archive dc_f668cc5ac03b, sha256 d45727496f86d1620ec90c60a19f84ba5ac52a8f1bb8f6104daf23d0e776e465)
BibTeX
@misc{dcf668cc5ac03b,
  title = {Sampling via Decision-Flow: Training-Free Extraction of Improved Latent Reasoning Paths in Large Language Models — arXiv:2609.12317},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2609.12317},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f668cc5ac03b; sha256 d45727496f86d1620ec90c60a19f84ba5ac52a8f1bb8f6104daf23d0e776e465},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.12317. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

A central question in LLM reasoning is whether reinforcement learning (RL) instills genuinely new capabilities or merely reshapes how existing knowledge is expressed during inference. Building on the distribution-sharpening hypothesis, which holds that RL reallocates probability mass toward high-reward trajectories already latent in base models, we ask: can we unlock those latent paths without costly RL fine-tuning? We present Decision-Flow Sampling (DF-Sample), a training-free, data-free inference-time framework that constructs a hierarchical reasoning tree, scores terminal nodes for quality, and back-propagates utilities to inform each intermediate branching decision. Unlike conventional sampling strategies that make purely local step-wise choices, DF-Sample performs explicit global trajectory evaluation before committing to a path, recovering high-quality but low-probability reasoning chains that standard decoding overlooks. On GPQA, DF-Sample achieves 45.6% accuracy, surpassing power sampling (38.9%) and GRPO (39.9%), showing that a training-free method can outperform a trained one. Across three models and four benchmarks, DF-Sample consistently outperforms baselines, indicating substantial latent reasoning potential in pretrained base models.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-14 in cs.LG).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2609.12317 — 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 2026-10-16T03:05:44.033Z dalla pagina https://arxiv.org/catchup/cs.LG/2026-09-14?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).