Home

Awesome

<div align="center"> <h1>Diffusion GLA (DiG) </h1> <h3>Scalable and Efficient Diffusion Models with Gated Linear Attention</h3>

Lianghui Zhu<sup>1,2</sup>,Zilong Huang<sup>2 :email:</sup>,Bencheng Liao<sup>1</sup>,Jun Hao Liew<sup>2</sup>, Hanshu Yan<sup>2</sup>, Jiashi Feng<sup>2</sup>, Xinggang Wang<sup>1 :email:</sup>

<sup>1</sup> School of EIC, Huazhong University of Science and Technology, <sup>2</sup> ByteDance

(<sup>:email:</sup>) corresponding author.

ArXiv Preprint (arXiv 2405.18428)

</div>

News

<div align="center"> <img src="assets/vis_teaser_v2.3.png" /> </div>

Abstract

Diffusion models with large-scale pre-training have achieved significant success in the field of visual content generation, particularly exemplified by Diffusion Transformers (DiT). However, DiT models have faced challenges with quadratic complexity efficiency, especially when handling long sequences. In this paper, we aim to incorporate the sub-quadratic modeling capability of Gated Linear Attention (GLA) into the 2D diffusion backbone. Specifically, we introduce Diffusion Gated Linear Attention Transformers (DiG), a simple, adoptable solution with minimal parameter overhead. We offer two variants, i,e, a plain and U-shape architecture, showing superior efficiency and competitive effectiveness. In addition to superior performance to DiT and other sub-quadratic-time diffusion models at $256 \times 256$ resolution, DiG demonstrates greater efficiency than these methods starting from a $512$ resolution. Specifically, DiG-S/2 is $2.5\times$ faster and saves $75.7%$ GPU memory compared to DiT-S/2 at a $1792$ resolution. Additionally, DiG-XL/2 is $4.2\times$ faster than the Mamba-based model at a $1024$ resolution and $1.8\times$ faster than DiT with FlashAttention-2 at a $2048$ resolution. We will release the code soon.

<div align="center"> <img src="assets/dig_teaser_v1.4.png" /> </div> <div align="center"> <img src="assets/scaling_err_v1.1.png" /> </div>

Overview

<div align="center"> <img src="assets/dig_pipeline_v3.6.png" /> </div>

Envs. for Training

Train Your DiG

Acknowledgement :heart:

This project is based on GLA (paper, code), flash-linear-attention (code), DiT (paper, code), DiS (paper, code), OpenDiT (code). Thanks for their wonderful works.

Citation

If you find DiG is useful in your research or applications, please consider giving us a star 🌟 and citing it by the following BibTeX entry.

@article{dig,
      title={DiG: Scalable and Efficient Diffusion Models with Gated Linear Attention}, 
      author={Lianghui Zhu and Zilong Huang and Bencheng Liao and Jun Hao Liew and Hanshu Yan and Jiashi Feng and Xinggang Wang},
      year={2024},
      eprint={2405.18428},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}