A Hitchhiker's Guide On Distributed Training of Deep Neural Networks

TL;DR

This paper surveys the state-of-the-art methods for distributed training of deep neural networks, highlighting techniques like synchronous/asynchronous SGD, gradient aggregation, and optimizations for high throughput and low latency.

Contribution

It provides a comprehensive overview of current algorithms and best practices for efficient distributed deep learning training frameworks.

Findings

Distributed training can significantly reduce training time for large datasets.

Techniques like mixed precision and gradient compression improve training efficiency.

State-of-the-art methods enable training on thousands of GPUs in minutes.

Abstract

Deep learning has led to tremendous advancements in the field of Artificial Intelligence. One caveat however is the substantial amount of compute needed to train these deep learning models. Training a benchmark dataset like ImageNet on a single machine with a modern GPU can take upto a week, distributing training on multiple machines has been observed to drastically bring this time down. Recent work has brought down ImageNet training time to a time as low as 4 minutes by using a cluster of 2048 GPUs. This paper surveys the various algorithms and techniques used to distribute training and presents the current state of the art for a modern distributed training framework. More specifically, we explore the synchronous and asynchronous variants of distributed Stochastic Gradient Descent, various All Reduce gradient aggregation strategies and best practices for obtaining higher throughout and lower latency over a cluster such as mixed precision training, large batch training and gradient compression.