Galvatron: Efficient Transformer Training over Multiple GPUs Using Automatic Parallelism
This addresses the problem of manual or limited parallelism planning for distributed training of large models, offering an incremental improvement in automation and efficiency.
The paper tackles the challenge of efficiently training Transformer models over multiple GPUs by proposing Galvatron, a system framework that automatically finds optimal hybrid parallelism strategies, achieving superior system throughput in evaluations across four workloads.
Transformer models have achieved state-of-the-art performance on various domains of applications and gradually becomes the foundations of the advanced large deep learning (DL) models. However, how to train these models over multiple GPUs efficiently is still challenging due to a large number of parallelism choices. Existing DL systems either rely on manual efforts to make distributed training plans or apply parallelism combinations within a very limited search space. In this approach, we propose Galvatron, a new system framework that incorporates multiple popular parallelism dimensions and automatically finds the most efficient hybrid parallelism strategy. To better explore such a rarely huge search space, we 1) involve a decision tree to make decomposition and pruning based on some reasonable intuitions, and then 2) design a dynamic programming search algorithm to generate the optimal plan. Evaluations on four representative Transformer workloads show that Galvatron could perform automatically distributed training with different GPU memory budgets. Among all evluated scenarios, Galvatron always achieves superior system throughput compared to previous work with limited parallelism.