Any-OPD：基于表示空间桥接的流匹配模型异构在线策略蒸馏

摘要

同策略蒸馏中，教师纠正学生自身生成的样本，这预设了两个模型使用相同的语言：相同的VAE潜变量、匹配的架构以及共同的时间步网格。我们探讨了当这些条件均不成立时会发生什么——例如可用的最强教师模型与希望部署的学生模型来自不同的模型家族——并发现标准方案对此无能为力：教师潜变量无法在异坐标系中充当目标；针对随机重绘局部细节的教师的逐像素损失会退化为模糊或发散；时间步索引在不匹配的调度中失去其意义。我们提出Any-OPD——据我们所知，这是首个支持任意潜变量流匹配生成器对之间进行同策略蒸馏的框架。Any-OPD将教师纯粹视为黑盒采样器，并仅在一点上连接两个模型：一个冻结的、与模型无关的视觉表示，在其中比较两者独立解码的输出，从而避开关于潜变量、特征或架构的所有假设。通过匹配连续噪声水平而非步索引来恢复轨迹对应关系；此外，一个简短的锚定阶段将教师样本通过学生自身的VAE重新编码，确保同策略梯度衡量的是样本质量而非域不匹配。将12B的FLUX.1-dev蒸馏至2.5B的SD3.5-Medium，Any-OPD将学生的PickScore从0.846提升到0.884，HPSv3从9.12提升到10.97，以五分之一规模的模型媲美教师，而直接潜变量回归在此场景下根本无法训练。

English

On-policy distillation, in which a teacher corrects samples that the student itself generates, presupposes that the two models speak the same language: identical VAE latents, matching architectures, and a common timestep grid. We ask what happens when none of this holds, as when the strongest teacher available and the student one wishes to deploy come from different model families, and find that the standard recipes have no answer: teacher latents cannot serve as targets in a foreign coordinate system, per-pixel losses against a teacher that stochastically re-draws local detail degenerate into blur or divergence, and timestep indices lose their meaning across mismatched schedules. We present Any-OPD, to our knowledge the first framework for on-policy distillation between arbitrary pairs of latent flow-matching generators. Any-OPD treats the teacher purely as a black-box sampler and connects the two models at exactly one point: a frozen, model-agnostic vision representation in which their independently decoded outputs are compared, sidestepping every assumption about latents, features, or architecture. Trajectory correspondence is recovered by matching continuous noise levels instead of step indices, and a brief anchoring phase, in which teacher samples are re-encoded through the student's own VAE, ensures the on-policy gradient measures sample quality rather than domain mismatch. Distilling the 12B FLUX.1-dev into the 2.5B SD3.5-Medium, Any-OPD lifts the student's PickScore from 0.846 to 0.884 and HPSv3 from 9.12 to 10.97, rivaling the teacher at a fifth of its size, where direct latent regression fails to train at all.