arXiv：cs.LG· Jinwoo Kim·· 3 小时前AI 评分30
约束感知训练：让语言模型在训练阶段跳过推理时已过滤的 token
Constraint-Aware Training
AI 导读
论文提出约束感知训练目标，将推理阶段通过程序分析过滤违规 token 的工作从训练中外部化，并给出关于模型规模与数据效率的三条定理。在受控合成实验中，相同参数量和数据下，约束感知训练比普通交叉熵训练取得更低的预测损失。该工作已被 Neurips 2026 workshop AI for Verifiable Coding 接收。
正文
Abstract:When generating programs with language models, constrained decoding can apply program analyses to exclude tokens that violate syntax, scope, or typing rules. However, there is a duplication: standard training already teaches the model to suppress the tokens rejected by these analyses. This duplication leads to the question: if we will perform some analysis to filter a set tokens out during inference anyways, can we avoid teaching the model the said analysis altogether during training, and does this externalization lead to more efficient models? This paper defines a general constraint-aware objective satisfying this externalization desideratum and formalizes the benefits of externalization into three concrete theorems about model size and data efficiency. We show, through a controlled synthetic experiment, that the theorems survive training dynamics: constraint-aware training yields lower prediction loss at a matched parameter count and data compared to ordinary cross-entropy training, motivating training objectives that incorporate the analyses used during generation.
| Comments: | Accepted at Neurips 2026 workshop AI for Verifiable Coding |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.02909 [cs.LG] |
| (or arXiv:2610.02909v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.02909 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jinwoo Kim [view email]
[v1]
Fri, 2 Oct 2026 06:59:13 UTC (57 KB)
来源：arXiv：cs.LG · arxiv.org