Saved in:
Bibliographic Details
Main Authors: Du, Zhenbang, Fu, Yonggan, Wang, Lifu, Qian, Jiayi, Luo, Xiao, Yingyan, Lin
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.06160
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916887654301696
author Du, Zhenbang
Fu, Yonggan
Wang, Lifu
Qian, Jiayi
Luo, Xiao
Yingyan
Lin
author_facet Du, Zhenbang
Fu, Yonggan
Wang, Lifu
Qian, Jiayi
Luo, Xiao
Yingyan
Lin
contents Diffusion models have shown remarkable success across generative tasks, yet their high computational demands challenge deployment on resource-limited platforms. This paper investigates a critical question for compute-optimal diffusion model deployment: Under a post-training setting without fine-tuning, is it more effective to reduce the number of denoising steps or to use a cheaper per-step inference? Intuitively, reducing the number of denoising steps increases the variability of the distributions across steps, making the model more sensitive to compression. In contrast, keeping more denoising steps makes the differences smaller, preserving redundancy, and making post-training compression more feasible. To systematically examine this, we propose PostDiff, a training-free framework for accelerating pre-trained diffusion models by reducing redundancy at both the input level and module level in a post-training manner. At the input level, we propose a mixed-resolution denoising scheme based on the insight that reducing generation resolution in early denoising steps can enhance low-frequency components and improve final generation fidelity. At the module level, we employ a hybrid module caching strategy to reuse computations across denoising steps. Extensive experiments and ablation studies demonstrate that (1) PostDiff can significantly improve the fidelity-efficiency trade-off of state-of-the-art diffusion models, and (2) to boost efficiency while maintaining decent generation fidelity, reducing per-step inference cost is often more effective than reducing the number of denoising steps. Our code is available at https://github.com/GATECH-EIC/PostDiff.
format Preprint
id arxiv_https___arxiv_org_abs_2508_06160
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Fewer Denoising Steps or Cheaper Per-Step Inference: Towards Compute-Optimal Diffusion Model Deployment
Du, Zhenbang
Fu, Yonggan
Wang, Lifu
Qian, Jiayi
Luo, Xiao
Yingyan
Lin
Computer Vision and Pattern Recognition
Diffusion models have shown remarkable success across generative tasks, yet their high computational demands challenge deployment on resource-limited platforms. This paper investigates a critical question for compute-optimal diffusion model deployment: Under a post-training setting without fine-tuning, is it more effective to reduce the number of denoising steps or to use a cheaper per-step inference? Intuitively, reducing the number of denoising steps increases the variability of the distributions across steps, making the model more sensitive to compression. In contrast, keeping more denoising steps makes the differences smaller, preserving redundancy, and making post-training compression more feasible. To systematically examine this, we propose PostDiff, a training-free framework for accelerating pre-trained diffusion models by reducing redundancy at both the input level and module level in a post-training manner. At the input level, we propose a mixed-resolution denoising scheme based on the insight that reducing generation resolution in early denoising steps can enhance low-frequency components and improve final generation fidelity. At the module level, we employ a hybrid module caching strategy to reuse computations across denoising steps. Extensive experiments and ablation studies demonstrate that (1) PostDiff can significantly improve the fidelity-efficiency trade-off of state-of-the-art diffusion models, and (2) to boost efficiency while maintaining decent generation fidelity, reducing per-step inference cost is often more effective than reducing the number of denoising steps. Our code is available at https://github.com/GATECH-EIC/PostDiff.
title Fewer Denoising Steps or Cheaper Per-Step Inference: Towards Compute-Optimal Diffusion Model Deployment
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.06160