FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yao, Jingfeng, Cheng, Wang, Liu, Wenyu, Wang, Xinggang
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917824544374784
author Yao, Jingfeng
Cheng, Wang
Liu, Wenyu
Wang, Xinggang
author_facet Yao, Jingfeng
Cheng, Wang
Liu, Wenyu
Wang, Xinggang
contents Diffusion Transformers (DiT) have attracted significant attention in research. However, they suffer from a slow convergence rate. In this paper, we aim to accelerate DiT training without any architectural modification. We identify the following issues in the training process: firstly, certain training strategies do not consistently perform well across different data. Secondly, the effectiveness of supervision at specific timesteps is limited. In response, we propose the following contributions: (1) We introduce a new perspective for interpreting the failure of the strategies. Specifically, we slightly extend the definition of Signal-to-Noise Ratio (SNR) and suggest observing the Probability Density Function (PDF) of SNR to understand the essence of the data robustness of the strategy. (2) We conduct numerous experiments and report over one hundred experimental results to empirically summarize a unified accelerating strategy from the perspective of PDF. (3) We develop a new supervision method that further accelerates the training process of DiT. Based on them, we propose FasterDiT, an exceedingly simple and practicable design strategy. With few lines of code modifications, it achieves 2.30 FID on ImageNet 256 resolution at 1000k iterations, which is comparable to DiT (2.27 FID) but 7 times faster in training.
format Preprint
id arxiv_https___arxiv_org_abs_2410_10356
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification
Yao, Jingfeng
Cheng, Wang
Liu, Wenyu
Wang, Xinggang
Computer Vision and Pattern Recognition
Diffusion Transformers (DiT) have attracted significant attention in research. However, they suffer from a slow convergence rate. In this paper, we aim to accelerate DiT training without any architectural modification. We identify the following issues in the training process: firstly, certain training strategies do not consistently perform well across different data. Secondly, the effectiveness of supervision at specific timesteps is limited. In response, we propose the following contributions: (1) We introduce a new perspective for interpreting the failure of the strategies. Specifically, we slightly extend the definition of Signal-to-Noise Ratio (SNR) and suggest observing the Probability Density Function (PDF) of SNR to understand the essence of the data robustness of the strategy. (2) We conduct numerous experiments and report over one hundred experimental results to empirically summarize a unified accelerating strategy from the perspective of PDF. (3) We develop a new supervision method that further accelerates the training process of DiT. Based on them, we propose FasterDiT, an exceedingly simple and practicable design strategy. With few lines of code modifications, it achieves 2.30 FID on ImageNet 256 resolution at 1000k iterations, which is comparable to DiT (2.27 FID) but 7 times faster in training.
title FasterDiT: Towards Faster Diffusion Transformers Training without Architecture Modification
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2410.10356