Improving Efficiency in Large-Scale Decentralized Distributed Training
This work addresses efficiency issues in large-scale decentralized distributed training for deep learning applications, offering incremental improvements to existing methods.
The paper tackles the problem of slow convergence in decentralized distributed training due to a decreasing spectral gap as the number of learners increases, and it proposes techniques to accelerate training by improving the spectral gap while minimizing communication cost, achieving the fastest reported training times on speech recognition and ImageNet tasks, such as training an LSTM acoustic model in 1.98 hours with 7.7% WER on Switchboard using 128 GPUs.
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.