Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Lewandowski, Basile, Kurz, Simon, Shankar, Aditya, Birke, Robert, Chen, Jian-Jia, Chen, Lydia Y.
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2603.14062
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866911515831959552
author Lewandowski, Basile
Kurz, Simon
Shankar, Aditya
Birke, Robert
Chen, Jian-Jia
Chen, Lydia Y.
author_facet Lewandowski, Basile
Kurz, Simon
Shankar, Aditya
Birke, Robert
Chen, Jian-Jia
Chen, Lydia Y.
contents Diffusion models are the go-to method for Text-to-Image generation, but their iterative denoising processes has high inference latency. Quantization reduces compute time by using lower bitwidths, but applies a fixed precision across all denoising timesteps, leaving an entire optimization axis unexplored. We propose TMPDiff, a temporal mixed-precision framework for diffusion models that assigns different numeric precision to different denoising timesteps. We hypothesize that quantization errors accumulate additively across timesteps, which we then validate experimentally. Based on our observations, we develop an adaptive bisectioning-based algorithm, which assigns per-step precisions with linear evaluation complexity, reducing an otherwise exponential search problem. Across four state-of-the-art diffusion models and three datasets, TMPDiff consistently outperforms uniform-precision baselines at matched speedup, achieving 10 to 20% improvement in perceptual quality. On FLUX.1-dev, TMPDiff achieves 90% SSIM relative to the full-precision model at a speedup of 2.5x over 16-bit inference.
format Preprint
id arxiv_https___arxiv_org_abs_2603_14062
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TMPDiff: Temporal Mixed-Precision for Diffusion Models
Lewandowski, Basile
Kurz, Simon
Shankar, Aditya
Birke, Robert
Chen, Jian-Jia
Chen, Lydia Y.
Computer Vision and Pattern Recognition
Machine Learning
Diffusion models are the go-to method for Text-to-Image generation, but their iterative denoising processes has high inference latency. Quantization reduces compute time by using lower bitwidths, but applies a fixed precision across all denoising timesteps, leaving an entire optimization axis unexplored. We propose TMPDiff, a temporal mixed-precision framework for diffusion models that assigns different numeric precision to different denoising timesteps. We hypothesize that quantization errors accumulate additively across timesteps, which we then validate experimentally. Based on our observations, we develop an adaptive bisectioning-based algorithm, which assigns per-step precisions with linear evaluation complexity, reducing an otherwise exponential search problem. Across four state-of-the-art diffusion models and three datasets, TMPDiff consistently outperforms uniform-precision baselines at matched speedup, achieving 10 to 20% improvement in perceptual quality. On FLUX.1-dev, TMPDiff achieves 90% SSIM relative to the full-precision model at a speedup of 2.5x over 16-bit inference.
title TMPDiff: Temporal Mixed-Precision for Diffusion Models
topic Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2603.14062