Improving Efficiency in Large-Scale Decentralized Distributed Training

TL;DR

This paper proposes techniques to accelerate decentralized distributed training algorithms by improving spectral gap, achieving faster training times for large-scale deep learning tasks on supercomputers.

Contribution

It introduces methods to enhance spectral gap in (A)D-PSGD, reducing communication costs and improving convergence speed in large-scale distributed training.

Findings

Achieved training an LSTM acoustic model in 2.28 hours on 64 GPUs.

Reached 7.5% WER on Switchboard test set.

Reported the fastest training times to date for large-scale tasks.

Abstract

Decentralized Parallel SGD (D-PSGD) and its asynchronous variant Asynchronous Parallel SGD (AD-PSGD) is a family of distributed learning algorithms that have been demonstrated to perform well for large-scale deep learning tasks. One drawback of (A)D-PSGD is that the spectral gap of the mixing matrix decreases when the number of learners in the system increases, which hampers convergence. In this paper, we investigate techniques to accelerate (A)D-PSGD based training by improving the spectral gap while minimizing the communication cost. We demonstrate the effectiveness of our proposed techniques by running experiments on the 2000-hour Switchboard speech recognition task and the ImageNet computer vision task. On an IBM P9 supercomputer, our system is able to train an LSTM acoustic model in 2.28 hours with 7.5% WER on the Hub5-2000 Switchboard (SWB) test set and 13.3% WER on the CallHome (CH) test set using 64 V100 GPUs and in 1.98 hours with 7.7% WER on SWB and 13.3% WER on CH using 128 V100 GPUs, the fastest training time reported to date.