STAC 改进分布匹配蒸馏：批评者模型需与教师同步扩大
A Better Critic Matters for Distribution Matching Distillation
这篇论文发现蒸馏时 critic 太小会把大教师带崩，提出的 STAC 按步条件化，在 SDXL、FLUX.2 和 Wan2.1 上都能提升少步生成质量，做蒸馏的值得看看。
Distribution Matching Distillation（DMD）依赖在线 critic 估计学生分布的 score 来训练少步生成模型。实验在 ImageNet 上覆盖 7 种 SiT 模型规模，发现只扩大教师而 critic 固定会导致学生质量下降甚至训练崩溃，同步扩大 critic（joint scaling）可避免崩溃。信息方面，作者提出 step-aware critic（STAC），按学生生成步数条件化估计 step-specific score。两项改进在 ImageNet 类条件生成、SDXL 与 FLUX.2 文生图、Wan2.1 文生视频上均带来一致的少步蒸馏提升。
A Better Critic Matters for Distribution Matching Distillation
Distribution Matching Distillation (DMD) trains a few-step student using the difference between a teacher's score and an online critic's estimate of the student distribution's score. We investigate what the critic needs, in capacity and in information, to provide a reliable training signal. On the capacity side, controlled experiments across seven SiT model sizes on ImageNet show that increasing teacher size while keeping a small critic fixed can degrade student quality and cause training collapse. Scaling the critic alongside the teacher, a strategy we call joint scaling, prevents collapse in our experiments and allows the student to benefit from larger teachers. On the information side, rather than forcing the critic to model a step-agnostic mixture of student outputs, we introduce a step-aware critic (STAC) that conditions on the student's generation step to estimate step-specific scores. Together, joint scaling and STAC consistently improve few-step distillation across class-conditional ImageNet generation, text-to-image generation with SDXL and FLUX.2, and text-to-video generation with Wan2.1.