Scalable and Efficient MoE Training for Multitask Multilingual Models

TL;DR

This paper introduces a scalable system and training methods for large sparse MoE models, enabling efficient training of trillion-parameter multilingual models that achieve state-of-the-art results in language tasks.

Contribution

The paper presents a system combining multi-dimensional parallelism and heterogeneous memory for efficient MoE training at trillions of parameters, along with new training techniques and expert pruning strategies.

Findings

Achieved 8x larger models on the same hardware compared to previous work.

Trained a 10-billion-parameter multilingual model with state-of-the-art performance.

Open-sourced the system implementation with DeepSpeed.

Abstract

The Mixture of Experts (MoE) models are an emerging class of sparsely activated deep learning models that have sublinear compute costs with respect to their parameters. In contrast with dense models, the sparse architecture of MoE offers opportunities for drastically growing model size with significant accuracy gain while consuming much lower compute budget. However, supporting large scale MoE training also has its own set of system and modeling challenges. To overcome the challenges and embrace the opportunities of MoE, we first develop a system capable of scaling MoE models efficiently to trillions of parameters. It combines multi-dimensional parallelism and heterogeneous memory technologies harmoniously with MoE to empower 8x larger models on the same hardware compared with existing work. Besides boosting system efficiency, we also present new training methods to improve MoE sample efficiency and leverage expert pruning strategy to improve inference time efficiency. By combining the efficient system and training methods, we are able to significantly scale up large multitask multilingual models for language generation which results in a great improvement in model accuracy. A model trained with 10 billion parameters on 50 languages can achieve state-of-the-art performance in Machine Translation (MT) and multilingual natural language generation tasks. The system support of efficient MoE training has been implemented and open-sourced with the DeepSpeed library.