发表机构
KU Leuven; Ruhr University Bochum（鲁汶大学; 波鸿鲁尔大学）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
该研究探究掩码扩散语言模型解码中无分类器引导（CFG）的必要性，定义承诺 horizon τ*，发现固定提示的交叉验证 horizon 时，其表现不劣于全程CFG，还可优化并行度与失败轨迹恢复效果。
AI 中文摘要
无分类器引导（CFG）通常在掩码扩散语言模型解码过程中全程保留，尽管其收益因提示和时间而异。我们通过比较任意部分输出在持续使用CFG和仅使用基础模型继续生成下，最终满足约束的概率，来研究何时真正需要CFG，二者的差值定义了引导的剩余价值。引导依赖具有高度提示特异性：许多提示无需CFG即可成功，而其他提示要么无显著收益，要么有害；对确实有收益的提示，增益通常集中在早期。我们将承诺 horizon τ*定义为：从该最早点起，将剩余所有解码切换为基础模型时，最终成功度降低不超过选定容差。在基础模型下，对应的成功概率（即承诺值）是鞅。一阶近似下，CFG的每步效应由引导logit方向与后继承诺值的协方差决定，这为引导何时有用提供了局部解释，但本身无法定位该 horizon。在观测到终端前 horizon 的提示中，τ*通常较早，且在约束族内的差异大于族间差异。在预设边际下，将每个提示固定在其自身交叉验证得到的 horizon 时，在全部13个子任务上的表现均不劣于全程使用CFG，即便仍有大量token被掩码。这将承诺与实现分离开来，该边界还识别出一个后期区域：更高的并行度仅会使约束成功度小幅下降，尽管流畅度仍随并行宽度降低。对于失败轨迹，重新开放已承诺的位置可改善两种失败模式的恢复效果。
英文摘要
Classifier-free guidance (CFG) is usually kept on throughout masked diffusion language model decoding, although its benefit varies across prompts and over time. We study when CFG is actually needed by comparing, from any partial output, the probability of eventual constraint satisfaction under continued CFG and under base-only continuation. Their difference defines the remaining value of guidance. Guidance dependence is highly prompt-specific. Many prompts already succeed without CFG, while for others it provides no measurable benefit or can be harmful. For prompts that do benefit, the gain is often concentrated early. We define the commitment horizon $\astar$ as the earliest point from which switching all remaining decoding to the base model reduces final success by no more than a chosen tolerance. Under the base model, the corresponding success probability, or committor, is a martingale. To first order, CFG's per-step effect is governed by the covariance between the guidance logit direction and the successor committor. This gives a local account of when guidance can help, but it does not by itself locate the horizon. Among prompts with an observed preterminal horizon, $\astar$ is usually early and varies more within constraint families than between them. Freezing each prompt at its own cross-fitted horizon is noninferior to full CFG on all 13 subtasks at the prespecified margin, even while many tokens remain masked. This separates commitment from realization. The boundary also identifies a later region in which higher parallelism adds only a small cost in constraint success, although fluency still degrades with parallel width. For failed trajectories, reopening committed positions improves recovery in both failure modes.
