EmbRace: Accelerating Sparse Communication for Distributed Training of NLP Neural Networks

TL;DR

EmbRace is a novel communication framework that significantly accelerates distributed training of sparse NLP models by reducing communication overhead through sparsity-aware hybrid communication and optimized scheduling.

Contribution

It introduces Sparsity-aware Hybrid Communication and 2D Communication Scheduling to improve scalability and efficiency in training sparse NLP models.

Findings

Achieves up to 2.41X speedup over baselines.

Effectively overlaps sparse communication with computation.

Reduces communication overhead for sparse parameters.

Abstract

Distributed data-parallel training has been widely adopted for deep neural network (DNN) models. Although current deep learning (DL) frameworks scale well for dense models like image classification models, we find that these DL frameworks have relatively low scalability for sparse models like natural language processing (NLP) models that have highly sparse embedding tables. Most existing works overlook the sparsity of model parameters thus suffering from significant but unnecessary communication overhead. In this paper, we propose EmbRace, an efficient communication framework to accelerate communications of distributed training for sparse models. EmbRace introduces Sparsity-aware Hybrid Communication, which integrates AlltoAll and model parallelism into data-parallel training, so as to reduce the communication overhead of highly sparse parameters. To effectively overlap sparse communication with both backward and forward computation, EmbRace further designs a 2D Communication Scheduling approach which optimizes the model computation procedure, relaxes the dependency of embeddings, and schedules the sparse communications of each embedding row with a priority queue. We have implemented a prototype of EmbRace based on PyTorch and Horovod, and conducted comprehensive evaluations with four representative NLP models. Experimental results show that EmbRace achieves up to 2.41X speedup compared to the state-of-the-art distributed training baselines.