Routing Matters in MoE: Scaling Diffusion Transformers with Explicit Routing Guidance

TL;DR

This paper introduces ProMoE, a novel MoE framework with explicit routing guidance for diffusion transformers, significantly improving expert specialization and performance in vision tasks like ImageNet classification.

Contribution

ProMoE employs a two-step routing process with conditional and prototypical routing, incorporating semantic guidance and a contrastive loss to enhance expert specialization in vision MoE.

Findings

ProMoE outperforms state-of-the-art methods on ImageNet.

Explicit routing guidance improves expert specialization.

Prototypical routing enhances intra-expert coherence and inter-expert diversity.

Abstract

Mixture-of-Experts (MoE) has emerged as a powerful paradigm for scaling model capacity while preserving computational efficiency. Despite its notable success in large language models (LLMs), existing attempts to apply MoE to Diffusion Transformers (DiTs) have yielded limited gains. We attribute this gap to fundamental differences between language and visual tokens. Language tokens are semantically dense with pronounced inter-token variation, while visual tokens exhibit spatial redundancy and functional heterogeneity, hindering expert specialization in vision MoE. To this end, we present ProMoE, an MoE framework featuring a two-step router with explicit routing guidance that promotes expert specialization. Specifically, this guidance encourages the router to partition image tokens into conditional and unconditional sets via conditional routing according to their functional roles, and refine the assignments of conditional image tokens through prototypical routing with learnable prototypes based on semantic content. Moreover, the similarity-based expert allocation in latent space enabled by prototypical routing offers a natural mechanism for incorporating explicit semantic guidance, and we validate that such guidance is crucial for vision MoE. Building on this, we propose a routing contrastive loss that explicitly enhances the prototypical routing process, promoting intra-expert coherence and inter-expert diversity. Extensive experiments on ImageNet benchmark demonstrate that ProMoE surpasses state-of-the-art methods under both Rectified Flow and DDPM training objectives. Code is available at https://github.com/ali-vilab/ProMoE.