擴散採樣的最優步長

Optimal Stepsize for Diffusion Sampling

March 27, 2025

作者: Jianning Pei, Han Hu, Shuyang Gu

cs.AI

摘要

擴散模型在生成質量上表現卓越，但由於次優的步長離散化，其採樣過程計算密集。現有研究主要集中於優化去噪方向，而我們則著眼於步長調度的原則性設計。本文提出最優步長蒸餾，這是一個動態規劃框架，通過從參考軌跡中提取知識來獲得理論上最優的調度方案。通過將步長優化重新表述為遞歸誤差最小化，我們的方法利用最優子結構特性，確保了全局離散化界限。關鍵在於，蒸餾出的調度方案在架構、ODE求解器和噪聲調度上展現出強大的魯棒性。實驗表明，文本到圖像生成速度提升了10倍，同時在GenEval上保持了99.4%的性能。我們的代碼可在https://github.com/bebebe666/OptimalSteps獲取。

English

Diffusion models achieve remarkable generation quality but suffer from computational intensive sampling due to suboptimal step discretization. While existing works focus on optimizing denoising directions, we address the principled design of stepsize schedules. This paper proposes Optimal Stepsize Distillation, a dynamic programming framework that extracts theoretically optimal schedules by distilling knowledge from reference trajectories. By reformulating stepsize optimization as recursive error minimization, our method guarantees global discretization bounds through optimal substructure exploitation. Crucially, the distilled schedules demonstrate strong robustness across architectures, ODE solvers, and noise schedules. Experiments show 10x accelerated text-to-image generation while preserving 99.4% performance on GenEval. Our code is available at https://github.com/bebebe666/OptimalSteps.