DiG: Scalable and Efficient Diffusion Models with Gated Linear Attention

TL;DR

This paper introduces DiG, a diffusion model with Gated Linear Attention that achieves higher efficiency and competitive performance for large-scale image generation, especially at high resolutions, with minimal parameter overhead.

Contribution

The paper presents DiG, a novel diffusion model incorporating Gated Linear Attention, offering sub-quadratic complexity, improved efficiency, and comparable effectiveness over existing models.

Findings

DiG outperforms DiT and other sub-quadratic diffusion models at 256x256 resolution.

DiG is 2.5x faster and uses 75.7% less GPU memory than DiT-S/2 at 1792 resolution.

DiG-XL/2 is 4.2x faster than Mamba and 1.8x faster than DiT with FlashAttention-2 at high resolutions.

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. Code is released at https://github.com/hustvl/DiG.