Taming Unbalanced Training Workloads in Deep Learning with Partial Collective Operations

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Li, Shigang, Ben-Nun, Tal, Di Girolamo, Salvatore, Alistarh, Dan, Hoefler, Torsten
Natura: Preprint
Pubblicazione: 2019
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909745552556032
author Li, Shigang
Ben-Nun, Tal
Di Girolamo, Salvatore
Alistarh, Dan
Hoefler, Torsten
author_facet Li, Shigang
Ben-Nun, Tal
Di Girolamo, Salvatore
Alistarh, Dan
Hoefler, Torsten
contents Load imbalance pervasively exists in distributed deep learning training systems, either caused by the inherent imbalance in learned tasks or by the system itself. Traditional synchronous Stochastic Gradient Descent (SGD) achieves good accuracy for a wide variety of tasks, but relies on global synchronization to accumulate the gradients at every training step. In this paper, we propose eager-SGD, which relaxes the global synchronization for decentralized accumulation. To implement eager-SGD, we propose to use two partial collectives: solo and majority. With solo allreduce, the faster processes contribute their gradients eagerly without waiting for the slower processes, whereas with majority allreduce, at least half of the participants must contribute gradients before continuing, all without using a central parameter server. We theoretically prove the convergence of the algorithms and describe the partial collectives in detail. Experimental results on load-imbalanced environments (CIFAR-10, ImageNet, and UCF101 datasets) show that eager-SGD achieves 1.27x speedup over the state-of-the-art synchronous SGD, without losing accuracy.
format Preprint
id arxiv_https___arxiv_org_abs_1908_04207
institution arXiv
publishDate 2019
record_format arxiv
spellingShingle Taming Unbalanced Training Workloads in Deep Learning with Partial Collective Operations
Li, Shigang
Ben-Nun, Tal
Di Girolamo, Salvatore
Alistarh, Dan
Hoefler, Torsten
Distributed, Parallel, and Cluster Computing
Machine Learning
Load imbalance pervasively exists in distributed deep learning training systems, either caused by the inherent imbalance in learned tasks or by the system itself. Traditional synchronous Stochastic Gradient Descent (SGD) achieves good accuracy for a wide variety of tasks, but relies on global synchronization to accumulate the gradients at every training step. In this paper, we propose eager-SGD, which relaxes the global synchronization for decentralized accumulation. To implement eager-SGD, we propose to use two partial collectives: solo and majority. With solo allreduce, the faster processes contribute their gradients eagerly without waiting for the slower processes, whereas with majority allreduce, at least half of the participants must contribute gradients before continuing, all without using a central parameter server. We theoretically prove the convergence of the algorithms and describe the partial collectives in detail. Experimental results on load-imbalanced environments (CIFAR-10, ImageNet, and UCF101 datasets) show that eager-SGD achieves 1.27x speedup over the state-of-the-art synchronous SGD, without losing accuracy.
title Taming Unbalanced Training Workloads in Deep Learning with Partial Collective Operations
topic Distributed, Parallel, and Cluster Computing
Machine Learning
url https://arxiv.org/abs/1908.04207