Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2020
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866918127999123456 |
|---|---|
| author | Li, Shigang Ben-Nun, Tal Nadiradze, Giorgi Di Girolamo, Salvatore Dryden, Nikoli Alistarh, Dan Hoefler, Torsten |
| author_facet | Li, Shigang Ben-Nun, Tal Nadiradze, Giorgi Di Girolamo, Salvatore Dryden, Nikoli Alistarh, Dan Hoefler, Torsten |
| contents | 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., 2.1x 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). |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2005_00124 |
| institution | arXiv |
| publishDate | 2020 |
| record_format | arxiv |
| spellingShingle | Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging Li, Shigang Ben-Nun, Tal Nadiradze, Giorgi Di Girolamo, Salvatore Dryden, Nikoli Alistarh, Dan Hoefler, Torsten Distributed, Parallel, and Cluster Computing Machine Learning C.1.4; D.1.3; I.2 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., 2.1x 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). |
| title | Breaking (Global) Barriers in Parallel Stochastic Optimization with Wait-Avoiding Group Averaging |
| topic | Distributed, Parallel, and Cluster Computing Machine Learning C.1.4; D.1.3; I.2 |
| url | https://arxiv.org/abs/2005.00124 |