cs.LG提交于 2026-10-08已译

噪声化提示(Noise Your Prompt):在连续扩散语言模型中对条件 Token 加噪

Noise Your Prompt: Noising Conditioning Tokens in Continuous Diffusion Language Models

Justin Jung

中文摘要

针对连续扩散语言模型文献中常见的做法,即在训练时保持条件提示 token 干净不变,本研究对该做法进行了重新审视,并提出一个非常简单的修改:在训练过程中也对条件提示 token 施加噪声。结果显示,采用这一修改后的训练目标后,模型在数独和 N 皇后等组合推理任务上获得了更好的泛化能力,其中在更困难的变体上提升最为显著(数独 Hard 的求解率从 3.73% 提升到 24.65%,10×10 N 皇后的解覆盖率从 50.60% 提升到 73.79%)。在自然语言生成方面,该方法在 Gigaword 摘要等中等规模数据集上带来了可测量的质量提升,但同时发现收益并不能迁移到所有自然语言任务(例如开放式对话生成)。所提方法仅对训练目标做单行修改,默认情况下不增加额外的推理开销,并且支持类无分类器引导(classifier-free guidance)式的引导采样。

关键要点

01问题:连续扩散语言模型训练时通常保持条件提示 token 干净,这可能限制了模型的泛化能力与生成多样性
02方法:对条件提示 token 同样施加噪声,仅修改一行训练目标,不增加默认推理开销,并支持类无分类器引导的引导采样
03结果:在组合推理任务上提升显著,数独 Hard 求解率从 3.73% 提升到 24.65%,10×10 N 皇后解覆盖率从 50.60% 提升到 73.79%
04结果:Gigaword 摘要等中等规模数据集上自然语言生成质量有所改善,但开放式对话生成等任务上未观察到收益迁移
05局限:对条件 token 加噪带来的改进无法泛化到所有自然语言任务,在开放式对话生成等场景中效果不明显

解读

尚无解读。

原始英文摘要

arXiv:2610.09145v1 Announce Type: new Abstract: We revisit a standard accepted practice in the continuous diffusion language model literature of fixing conditioning prompt tokens clean during training. We make a very simple modification: also noise the conditioning prompt tokens during training. We demonstrate that under this modified training objective, we achieve better generalization in combinatorial reasoning tasks such as Sudoku and N-Queens, with the largest gains on harder variants ($3.73\% \to 24.65\%$ solve rate on Sudoku Hard), and increased diversity of generated solutions ($50.60\% \to 73.79\%$ coverage on 10x10 N-Queens). We also show measurable improvements to natural language generation quality in modest dataset regimes with Gigaword summarization, but notably demonstrate that gains do not transfer to all natural language tasks (e.g open ended dialogue generation). Our method is a single line change to the training objective, requires no additional inference costs by default, and provides the flexibility of classifier-free guidance inspired guided sampling. Our \href{https://github.com/LateralIntelligence/noise-your-prompt} {code} is publicly available.