解耦DMD：CFG增强作为矛，分布匹配作为盾
Decoupled DMD: CFG Augmentation as the Spear, Distribution Matching as the Shield
AI总结：
本文提出解耦DMD方法，发现CFG增强是蒸馏核心驱动因素，分布匹配作为正则化器，通过解耦噪声调度提升生成性能。
摘要：
扩散模型蒸馏已经发展成为一种强大的技术，用于创建高效的少步和单步生成器。其中，分布匹配蒸馏（DMD）及其变体因其出色的性能而脱颖而出，其性能通常归因于其核心机制，即使学生模型的输出分布与预训练教师模型的输出分布相匹配。在本文中，我们挑战了这种传统理解。通过严格分解DMD训练目标，我们发现，在像文本到图像生成这样的复杂任务中，其中CFG通常用于获得理想的少步性能，少步蒸馏的主要驱动力不是分布匹配，而是一个之前被忽视的组件，我们将其称为CFG增强（CA）。我们证明了这一术语作为核心“引擎”蒸馏的作用，而分布匹配（DM）术语则作为“正则化器”，确保训练稳定性并减轻伪影。我们进一步验证了这种解耦，通过证明虽然DM项是一个非常有效的正则化器，但它并不唯一；更简单的非参数约束或基于GAN的目标可以起到相同稳定作用，尽管有不同权衡。这种解耦劳动促使对两个术语性质的更系统分析，从而导致更系统和深入的理解。这种新理解进一步使我们能够提出对蒸馏过程的有原则修改，例如解耦引擎和正则化器的噪声调度，从而进一步提升性能。值得注意的是，我们的方法已被Z-Image（https://github.com/Tongyi-MAI/Z-Image）项目采用，以开发一个顶级8步图像生成模型，实证验证了我们发现的泛化性和鲁棒性。
英文摘要：
Diffusion model distillation has emerged as a powerful technique for creating efficient few-step and single-step generators. Among these, Distribution Matching Distillation (DMD) and its variants stand out for their impressive performance, which is widely attributed to their core mechanism of matching the student's output distribution to that of a pre-trained teacher model. In this work, we challenge this conventional understanding. Through a rigorous decomposition of the DMD training objective, we reveal that in complex tasks like text-to-image generation, where CFG is typically required for desirable few-step performance, the primary driver of few-step distillation is not distribution matching, but a previously overlooked component we identify as CFG Augmentation (CA). We demonstrate that this term acts as the core ``engine'' of distillation, while the Distribution Matching (DM) term functions as a ``regularizer'' that ensures training stability and mitigates artifacts. We further validate this decoupling by demonstrating that while the DM term is a highly effective regularizer, it is not unique; simpler non-parametric constraints or GAN-based objectives can serve the same stabilizing function, albeit with different trade-offs. This decoupling of labor motivates a more principled analysis of the properties of both terms, leading to a more systematic and in-depth understanding. This new understanding further enables us to propose principled modifications to the distillation process, such as decoupling the noise schedules for the engine and the regularizer, leading to further performance gains. Notably, our method has been adopted by the Z-Image ( https://github.com/Tongyi-MAI/Z-Image ) project to develop a top-tier 8-step image generation model, empirically validating the generalization and robustness of our findings.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
