Optimal Stepsize for Diffusion Sampling
This addresses the slow sampling problem in diffusion models for generative AI applications, representing a novel method for a known bottleneck rather than an incremental improvement.
The paper tackles the computational inefficiency of diffusion model sampling by proposing Optimal Stepsize Distillation, a dynamic programming framework that extracts theoretically optimal stepsize schedules, achieving 10x accelerated text-to-image generation while preserving 99.4% performance on GenEval.
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.