AutoDDL: Automatic Distributed Deep Learning with Near-Optimal Bandwidth Cost

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Chen, Jinfan, Li, Shigang, Gun, Ran, Yuan, Jinhui, Hoefler, Torsten
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916910274183168
author Chen, Jinfan
Li, Shigang
Gun, Ran
Yuan, Jinhui
Hoefler, Torsten
author_facet Chen, Jinfan
Li, Shigang
Gun, Ran
Yuan, Jinhui
Hoefler, Torsten
contents Recent advances in deep learning are driven by the growing scale of computation, data, and models. However, efficiently training large-scale models on distributed systems requires an intricate combination of data, operator, and pipeline parallelism, which exerts heavy burden on machine learning practitioners. To this end, we propose AutoDDL, a distributed training framework that automatically explores and exploits new parallelization schemes with near-optimal bandwidth cost. AutoDDL facilitates the description and implementation of different schemes by utilizing OneFlow's Split, Broadcast, and Partial Sum (SBP) abstraction. AutoDDL is equipped with an analytical performance model combined with a customized Coordinate Descent algorithm, which significantly reduces the scheme searching overhead. We conduct evaluations on Multi-Node-Single-GPU and Multi-Node-Multi-GPU machines using different models, including VGG and Transformer. Compared to the expert-optimized implementations, AutoDDL reduces the end-to-end training time by up to 31.1% and 10% for Transformer and up to 17.7% and 71.5% for VGG on the two parallel systems, respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2301_06813
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle AutoDDL: Automatic Distributed Deep Learning with Near-Optimal Bandwidth Cost
Chen, Jinfan
Li, Shigang
Gun, Ran
Yuan, Jinhui
Hoefler, Torsten
Distributed, Parallel, and Cluster Computing
C.1.4; I.2.11
Recent advances in deep learning are driven by the growing scale of computation, data, and models. However, efficiently training large-scale models on distributed systems requires an intricate combination of data, operator, and pipeline parallelism, which exerts heavy burden on machine learning practitioners. To this end, we propose AutoDDL, a distributed training framework that automatically explores and exploits new parallelization schemes with near-optimal bandwidth cost. AutoDDL facilitates the description and implementation of different schemes by utilizing OneFlow's Split, Broadcast, and Partial Sum (SBP) abstraction. AutoDDL is equipped with an analytical performance model combined with a customized Coordinate Descent algorithm, which significantly reduces the scheme searching overhead. We conduct evaluations on Multi-Node-Single-GPU and Multi-Node-Multi-GPU machines using different models, including VGG and Transformer. Compared to the expert-optimized implementations, AutoDDL reduces the end-to-end training time by up to 31.1% and 10% for Transformer and up to 17.7% and 71.5% for VGG on the two parallel systems, respectively.
title AutoDDL: Automatic Distributed Deep Learning with Near-Optimal Bandwidth Cost
topic Distributed, Parallel, and Cluster Computing
C.1.4; I.2.11
url https://arxiv.org/abs/2301.06813