Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging
Abstract
Deep learning at scale is dominated by communication time. Distributing samples across nodes usually yields the best performance, but poses scaling challenges due to global information dissemination and load imbalance across uneven sample lengths. State-of-the-art decentralized optimizers mitigate the problem, but require more iterations to achieve the same accuracy as their globally-communicating counterparts. We present Wait-Avoiding Group Model Averaging (WAGMA) SGD, a wait-avoiding stochastic optimizer that reduces global communication via subgroup weight exchange. The key insight is a combination of algorithmic changes to the averaging scheme and the use of a group allreduce operation. We prove the convergence of WAGMA-SGD, and empirically show that it retains convergence rates similar to Allreduce-SGD. For evaluation, we train ResNet-50 on ImageNet; Transformer for machine translation; and deep reinforcement learning for navigation at scale. Compared with state-of-the-art decentralized SGD variants, WAGMA-SGD significantly improves training throughput (e.g., 2.1× on 1,024 GPUs for reinforcement learning), and achieves the fastest time-to-solution (e.g., the highest score using the shortest training time for Transformer). © 1990-2012 IEEE. Show more
Publication status
publishedExternal links
Journal / series
IEEE Transactions on Parallel and Distributed SystemsVolume
Pages / Article No.
Publisher
IEEESubject
Stochastic gradient descent; distributed deep learning; decentralized optimizationOrganisational unit
03950 - Hoefler, Torsten / Hoefler, Torsten
Funding
678880 - DAPP: Data-Centric Parallel Programming (EC)
801039 - Exascale Programming Models for Heterogeneous Systems (EC)
185778 - Empowering Computational Science using Data-Centric Programming (SNF)
More
Show all metadata