Optimal Stepsize for Diffusion Sampling

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Pei, Jianning, Hu, Han, Gu, Shuyang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909555425804288
author Pei, Jianning
Hu, Han
Gu, Shuyang
author_facet Pei, Jianning
Hu, Han
Gu, Shuyang
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2503_21774
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Optimal Stepsize for Diffusion Sampling
Pei, Jianning
Hu, Han
Gu, Shuyang
Computer Vision and Pattern Recognition
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.
title Optimal Stepsize for Diffusion Sampling
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2503.21774