OSDP: Optimal Sharded Data Parallel for Distributed Deep Learning

TL;DR

OSDP is an automated system that optimally combines data and model parallelism for distributed deep learning, improving training efficiency and enabling larger models with higher throughput.

Contribution

It introduces OSDP, a novel automated parallel training system that balances memory and hardware utilization, and employs operator splitting to reduce peak memory usage during training.

Findings

Outperforms state-of-the-art methods in multiple large-scale models

Enables training of larger models with higher throughput

Demonstrates significant efficiency improvements in distributed training

Abstract

Large-scale deep learning models contribute to significant performance improvements on varieties of downstream tasks. Current data and model parallelism approaches utilize model replication and partition techniques to support the distributed training of ultra-large models. However, directly deploying these systems often leads to sub-optimal training efficiency due to the complex model architectures and the strict device memory constraints. In this paper, we propose Optimal Sharded Data Parallel (OSDP), an automated parallel training system that combines the advantages from both data and model parallelism. Given the model description and the device information, OSDP makes trade-offs between the memory consumption and the hardware utilization, thus automatically generates the distributed computation graph and maximizes the overall system throughput. In addition, OSDP introduces operator splitting to further alleviate peak memory footprints during training with negligible overheads, which enables the trainability of larger models as well as the higher throughput. Extensive experimental results of OSDP on multiple different kinds of large-scale models demonstrate that the proposed strategy outperforms the state-of-the-art in multiple regards. Our code is available at https://github.com/Youhe-Jiang/OptimalShardedDataParallel.