Abstract:Reinforcement Learning (RL) post-training alignment for language models is effective, but also costly and unstable in practice, owing to its complicated training process. To address this, we propose a training-free inference method to sample directly from the optimal RL policy. The transition probability applied to Masked Language Modeling (MLM) consists of a reference policy model and an energy term. Based on this, our algorithm, Energy-Guided Test-Time Scaling (ETS), estimates the key energy term via online Monte Carlo, with a provable convergence rate. Moreover, to ensure practical efficiency, ETS leverages modern acceleration frameworks alongside tailored importance sampling estimators, substantially reducing inference latency while provably preserving sampling quality. Experiments on MLM (including autoregressive models and diffusion language models) across reasoning, coding, and science benchmarks show that our ETS consistently improves generation quality, validating its effectiveness and design. The code is available at this https URL.
| Comments: | Accepted by ICML 2026 |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2601.21484 [cs.LG] |
| (or arXiv:2601.21484v3 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2601.21484 arXiv-issued DOI via DataCite |
From: Xiuyu Li [view email]
[v1]
Thu, 29 Jan 2026 10:06:52 UTC (1,373 KB)
[v2]
Sat, 9 May 2026 02:30:01 UTC (1,382 KB)
[v3]
Tue, 19 May 2026 09:15:43 UTC (1,380 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。