Diffusion models have emerged as mainstream framework in visual generation. Building upon this success, the integration of Mixture of Experts (MoE) methods has shown promise in enhancing model scalability and performance. In this paper, we introduce Race-DiT, a novel MoE model for diffusion transformers with a flexible routing strategy, Expert Race. By allowing tokens and experts to compete together and select the top candidates, the model learns to dynamically assign experts to critical tokens. Additionally, we propose per-layer regularization to address challenges in shallow layer learning, and router similarity loss to prevent mode collapse, ensuring better expert utilization. Extensive experiments on ImageNet validate the effectiveness of our approach, showcasing significant performance gains while promising scaling properties.
View on arXiv@article{yuan2025_2503.16057,
title={ Expert Race: A Flexible Routing Strategy for Scaling Diffusion Transformer with Mixture of Experts },
author={ Yike Yuan and Ziyu Wang and Zihao Huang and Defa Zhu and Xun Zhou and Jingyi Yu and Qiyang Min },
journal={arXiv preprint arXiv:2503.16057},
year={ 2025 }
}