MPDiT: Multi-Patch Global-to-Local Transformer Architecture For Efficient Flow Matching and Diffusion Model

TL;DR

MPDiT introduces a hierarchical multi-patch transformer architecture that reduces computational costs by processing larger patches early on and smaller patches later, while maintaining strong generative performance.

Contribution

The paper proposes a novel multi-patch transformer design with hierarchical patch sizes and improved embeddings, enhancing efficiency and training speed in diffusion models.

Findings

Reduces GFLOPs by up to 50% compared to isotropic DiTs.

Achieves comparable or better generative performance on ImageNet.

Accelerates training convergence with improved embeddings.

Abstract

Transformer architectures, particularly Diffusion Transformers (DiTs), have become widely used in diffusion and flow-matching models due to their strong performance compared to convolutional UNets. However, the isotropic design of DiTs processes the same number of patchified tokens in every block, leading to relatively heavy computation during training process. In this work, we introduce a multi-patch transformer design in which early blocks operate on larger patches to capture coarse global context, while later blocks use smaller patches to refine local details. This hierarchical design could reduces computational cost by up to 50% in GFLOPs while achieving good generative performance. In addition, we also propose improved designs for time and class embeddings that accelerate training convergence. Extensive experiments on the ImageNet dataset demonstrate the effectiveness of our architectural choices. Code is released at: https://github.com/quandao10/MPDiT