Скачать или смотреть OSDI '25 - ZEN: Empowering Distributed Training with Sparsity-driven Data Synchronization

ZEN: Empowering Distributed Training with Sparsity-driven Data Synchronization

Zhuang Wang, Rice University; Zhaozhuo Xu, Stevens Institute of Technology; Jingyi Xi, unaffiliated; Yuke Wang, Anshumali Shrivastava, and T. S. Eugene Ng, Rice University

Distributed training is the de facto standard to scale up the training of deep learning models with multiple GPUs. Its performance bottleneck lies in communications for gradient synchronization. Although high tensor sparsity is widely observed, the optimal communication scheme to fully leverage sparsity is still missing. This paper aims to bridge this gap. We first analyze the characteristics of sparse tensors in popular models to understand the fundamentals of sparsity. We then systematically explore the design space of communication schemes for sparse tensors and find the optimal ones. These findings give a new understanding and inspire us to develop a holistic gradient synchronization system for sparse tensors called ZEN. We demonstrate that ZEN can achieve up to 5.09x speedup in communication time and up to 2.48x speedup in training throughput compared to the state-of-the-art methods.

View the full OSDI '25 program at https://www.usenix.org/conference/osd...