Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging

Li S, Tal Ben-Nun TB-N, Nadiradze G, Girolamo SD, Dryden N, Alistarh D-A, Hoefler T. 2020. Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging. IEEE Transactions on Parallel and Distributed Systems., 3040606.


Journal Article | Epub ahead of print | English

Scopus indexed
Author
Li, Shigang; Tal Ben-Nun, Tal Ben-Nun; Nadiradze, GiorgiIST Austria; Girolamo, Salvatore Di; Dryden, Nikoli; Alistarh, Dan-AdrianIST Austria; Hoefler, Torsten
Department
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 equivalent 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., 2.1x 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).
Publishing Year
Date Published
2020-11-25
Journal Title
IEEE Transactions on Parallel and Distributed Systems
Article Number
3040606
ISSN
IST-REx-ID

Cite this

Li S, Tal Ben-Nun TB-N, Nadiradze G, et al. Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging. IEEE Transactions on Parallel and Distributed Systems. 2020. doi:10.1109/TPDS.2020.3040606
Li, S., Tal Ben-Nun, T. B.-N., Nadiradze, G., Girolamo, S. D., Dryden, N., Alistarh, D.-A., & Hoefler, T. (2020). Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging. IEEE Transactions on Parallel and Distributed Systems. IEEE. https://doi.org/10.1109/TPDS.2020.3040606
Li, Shigang, Tal Ben-Nun Tal Ben-Nun, Giorgi Nadiradze, Salvatore Di Girolamo, Nikoli Dryden, Dan-Adrian Alistarh, and Torsten Hoefler. “Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging.” IEEE Transactions on Parallel and Distributed Systems. IEEE, 2020. https://doi.org/10.1109/TPDS.2020.3040606.
S. Li et al., “Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging,” IEEE Transactions on Parallel and Distributed Systems. IEEE, 2020.
Li S, Tal Ben-Nun TB-N, Nadiradze G, Girolamo SD, Dryden N, Alistarh D-A, Hoefler T. 2020. Breaking (global) barriers in parallel stochastic optimization with wait-avoiding group averaging. IEEE Transactions on Parallel and Distributed Systems., 3040606.
Li, Shigang, et al. “Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging.” IEEE Transactions on Parallel and Distributed Systems, 3040606, IEEE, 2020, doi:10.1109/TPDS.2020.3040606.
All files available under the following license(s):
Copyright Statement:
This Item is protected by copyright and/or related rights. [...]

Link(s) to Main File(s)
Access Level
OA Open Access

Export

Marked Publications

Open Data IST Research Explorer

Sources

arXiv 2005.00124

Search this title in

Google Scholar