Enregistré dans:
Détails bibliographiques
Auteurs principaux: Jiang, Ting, Wang, Yixiao, Ye, Hancheng, Shao, Zishan, Sun, Jingwei, Zhang, Jingyang, Chen, Zekai, Zhang, Jianyi, Chen, Yiran, Li, Hai
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2507.17135
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866913955765551104
author Jiang, Ting
Wang, Yixiao
Ye, Hancheng
Shao, Zishan
Sun, Jingwei
Zhang, Jingyang
Chen, Zekai
Zhang, Jianyi
Chen, Yiran
Li, Hai
author_facet Jiang, Ting
Wang, Yixiao
Ye, Hancheng
Shao, Zishan
Sun, Jingwei
Zhang, Jingyang
Chen, Zekai
Zhang, Jianyi
Chen, Yiran
Li, Hai
contents Diffusion models have achieved remarkable success in generative tasks but suffer from high computational costs due to their iterative sampling process and quadratic attention costs. Existing training-free acceleration strategies that reduce per-step computation cost, while effectively reducing sampling time, demonstrate low faithfulness compared to the original baseline. We hypothesize that this fidelity gap arises because (a) different prompts correspond to varying denoising trajectory, and (b) such methods do not consider the underlying ODE formulation and its numerical solution. In this paper, we propose Stability-guided Adaptive Diffusion Acceleration (SADA), a novel paradigm that unifies step-wise and token-wise sparsity decisions via a single stability criterion to accelerate sampling of ODE-based generative models (Diffusion and Flow-matching). For (a), SADA adaptively allocates sparsity based on the sampling trajectory. For (b), SADA introduces principled approximation schemes that leverage the precise gradient information from the numerical ODE solver. Comprehensive evaluations on SD-2, SDXL, and Flux using both EDM and DPM++ solvers reveal consistent $\ge 1.8\times$ speedups with minimal fidelity degradation (LPIPS $\leq 0.10$ and FID $\leq 4.5$) compared to unmodified baselines, significantly outperforming prior methods. Moreover, SADA adapts seamlessly to other pipelines and modalities: It accelerates ControlNet without any modifications and speeds up MusicLDM by $1.8\times$ with $\sim 0.01$ spectrogram LPIPS.
format Preprint
id arxiv_https___arxiv_org_abs_2507_17135
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SADA: Stability-guided Adaptive Diffusion Acceleration
Jiang, Ting
Wang, Yixiao
Ye, Hancheng
Shao, Zishan
Sun, Jingwei
Zhang, Jingyang
Chen, Zekai
Zhang, Jianyi
Chen, Yiran
Li, Hai
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Diffusion models have achieved remarkable success in generative tasks but suffer from high computational costs due to their iterative sampling process and quadratic attention costs. Existing training-free acceleration strategies that reduce per-step computation cost, while effectively reducing sampling time, demonstrate low faithfulness compared to the original baseline. We hypothesize that this fidelity gap arises because (a) different prompts correspond to varying denoising trajectory, and (b) such methods do not consider the underlying ODE formulation and its numerical solution. In this paper, we propose Stability-guided Adaptive Diffusion Acceleration (SADA), a novel paradigm that unifies step-wise and token-wise sparsity decisions via a single stability criterion to accelerate sampling of ODE-based generative models (Diffusion and Flow-matching). For (a), SADA adaptively allocates sparsity based on the sampling trajectory. For (b), SADA introduces principled approximation schemes that leverage the precise gradient information from the numerical ODE solver. Comprehensive evaluations on SD-2, SDXL, and Flux using both EDM and DPM++ solvers reveal consistent $\ge 1.8\times$ speedups with minimal fidelity degradation (LPIPS $\leq 0.10$ and FID $\leq 4.5$) compared to unmodified baselines, significantly outperforming prior methods. Moreover, SADA adapts seamlessly to other pipelines and modalities: It accelerates ControlNet without any modifications and speeds up MusicLDM by $1.8\times$ with $\sim 0.01$ spectrogram LPIPS.
title SADA: Stability-guided Adaptive Diffusion Acceleration
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2507.17135