通过REINFORCE与James-Stein收缩设计实例级采样时间表
Designing Instance-Level Sampling Schedules via REINFORCE with James-Stein Shrinkage
提出一种模型无关的后训练方法，通过单次Dirichlet策略学习实例级采样时间表，并利用James-Stein估计器构建奖励基线以降低梯度方差，可即插即用地提升冻结文本到图像扩散模型的文本对齐和少步生成质量。
原文依据 we learn instance-level (prompt- and noise-conditioned) schedules through a single-pass Dirichlet policy
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论大多数文本到图像采样器的后训练方法聚焦于模型权重：要么微调骨干网络以实现对齐，要么蒸馏以实现少步效率。我们采取不同的路径：重新调度冻结采样器的采样时间线。我们不是使用固定的全局时间表，而是通过单次Dirichlet策略学习实例级（以提示和噪声为条件）的时间表。为了在高维策略学习中确保准确的梯度估计，我们引入了一种基于有原则的James-Stein估计器的新颖奖励基线；它可证明地比常用变体实现更低的估计误差，并带来更优的结果。我们重新调度的采样器在现代Stable Diffusion和Flux模型系列中持续改进了文本-图像对齐，包括文本渲染和组合控制。此外，采用我们时间表的5步Flux-Dev采样器可以达到与专门蒸馏的采样器（如Flux-Schnell）相当的生成质量。因此，我们将我们的调度框架定位为一种新兴的模型无关后训练杠杆，释放预训练采样器的额外生成潜力。
英文原文摘要 1144 字符
Most post-training methods for text-to-image samplers focus on the model weights: either fine-tuning the backbone for alignment or distilling it for few-step efficiency. We take a different route: rescheduling the sampling timeline of a frozen sampler. Instead of a fixed, global schedule, we learn instance-level (prompt- and noise-conditioned) schedules through a single-pass Dirichlet policy. To ensure accurate gradient estimates in high-dimensional policy learning, we introduce a novel reward baseline based on a principled James-Stein estimator; it provably achieves lower estimation errors than commonly used variants and leads to superior results. Our rescheduled samplers consistently improve text-image alignment including text rendering and compositional control across modern Stable Diffusion and Flux model families.Additionally, a 5-step Flux-Dev sampler with our schedules can attain generation quality comparable to deliberately distilled samplers like Flux-Schnell. We thus position our scheduling framework as an emerging model-agnostic post-training lever that unlocks additional generative potential in pretrained samplers.