FREE: Uncertainty-Aware Autoregression for Parallel Diffusion Transformers

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Wen, Xinwan, Li, Bowen, Luo, Jiajun, Li, Ye, Wang, Zhi
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866908675480748032
author Wen, Xinwan
Li, Bowen
Luo, Jiajun
Li, Ye
Wang, Zhi
author_facet Wen, Xinwan
Li, Bowen
Luo, Jiajun
Li, Ye
Wang, Zhi
contents Diffusion Transformers (DiTs) achieve state-of-the-art generation quality but require long sequential denoising trajectories, leading to high inference latency. Recent speculative inference methods enable lossless parallel sampling in U-Net-based diffusion models via a drafter-verifier scheme, but their acceleration is limited on DiTs due to insufficient draft accuracy during verification. To address this limitation, we analyze the DiTs' feature dynamics and find the features of the final transformer layer (top-block) exhibit strong temporal consistency and rich semantic abstraction. Based on this insight, we propose FREE, a novel framework that employs a lightweight drafter to perform feature-level autoregression with parallel verification, guaranteeing lossless acceleration with theoretical and empirical support. Meanwhile, prediction variance (uncertainty) of DiTs naturally increases in later denoising steps, reducing acceptance rates under speculative sampling. To mitigate this effect, we further introduce an uncertainty-guided relaxation strategy, forming FREE (relax), which dynamically adjusts the acceptance probability in response to uncertainty levels. Experiments on ImageNet-$512^2$ show that FREE achieves up to $1.86 \times$ acceleration, and FREE (relax) further reaches $2.25 \times$ speedup while maintaining high perceptual and quantitative fidelity in generation quality.
format Preprint
id arxiv_https___arxiv_org_abs_2511_20390
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle FREE: Uncertainty-Aware Autoregression for Parallel Diffusion Transformers
Wen, Xinwan
Li, Bowen
Luo, Jiajun
Li, Ye
Wang, Zhi
Computer Vision and Pattern Recognition
Diffusion Transformers (DiTs) achieve state-of-the-art generation quality but require long sequential denoising trajectories, leading to high inference latency. Recent speculative inference methods enable lossless parallel sampling in U-Net-based diffusion models via a drafter-verifier scheme, but their acceleration is limited on DiTs due to insufficient draft accuracy during verification. To address this limitation, we analyze the DiTs' feature dynamics and find the features of the final transformer layer (top-block) exhibit strong temporal consistency and rich semantic abstraction. Based on this insight, we propose FREE, a novel framework that employs a lightweight drafter to perform feature-level autoregression with parallel verification, guaranteeing lossless acceleration with theoretical and empirical support. Meanwhile, prediction variance (uncertainty) of DiTs naturally increases in later denoising steps, reducing acceptance rates under speculative sampling. To mitigate this effect, we further introduce an uncertainty-guided relaxation strategy, forming FREE (relax), which dynamically adjusts the acceptance probability in response to uncertainty levels. Experiments on ImageNet-$512^2$ show that FREE achieves up to $1.86 \times$ acceleration, and FREE (relax) further reaches $2.25 \times$ speedup while maintaining high perceptual and quantitative fidelity in generation quality.
title FREE: Uncertainty-Aware Autoregression for Parallel Diffusion Transformers
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2511.20390