FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tang, Zhenheng, Kang, Xueze, Yin, Yiming, Pan, Xinglin, Wang, Yuxin, He, Xin, Wang, Qiang, Zeng, Rongfei, Zhao, Kaiyong, Shi, Shaohuai, Zhou, Amelie Chi, Li, Bo, He, Bingsheng, Chu, Xiaowen
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909352052391936
author Tang, Zhenheng
Kang, Xueze
Yin, Yiming
Pan, Xinglin
Wang, Yuxin
He, Xin
Wang, Qiang
Zeng, Rongfei
Zhao, Kaiyong
Shi, Shaohuai
Zhou, Amelie Chi
Li, Bo
He, Bingsheng
Chu, Xiaowen
author_facet Tang, Zhenheng
Kang, Xueze
Yin, Yiming
Pan, Xinglin
Wang, Yuxin
He, Xin
Wang, Qiang
Zeng, Rongfei
Zhao, Kaiyong
Shi, Shaohuai
Zhou, Amelie Chi
Li, Bo
He, Bingsheng
Chu, Xiaowen
contents To alleviate hardware scarcity in training large deep neural networks (DNNs), particularly large language models (LLMs), we present FusionLLM, a decentralized training system designed and implemented for training DNNs using geo-distributed GPUs across different computing clusters or individual devices. Decentralized training faces significant challenges regarding system design and efficiency, including: 1) the need for remote automatic differentiation (RAD), 2) support for flexible model definitions and heterogeneous software, 3) heterogeneous hardware leading to low resource utilization or the straggler problem, and 4) slow network communication. To address these challenges, in the system design, we represent the model as a directed acyclic graph of operators (OP-DAG). Each node in the DAG represents the operator in the DNNs, while the edge represents the data dependency between operators. Based on this design, 1) users are allowed to customize any DNN without caring low-level operator implementation; 2) we enable the task scheduling with the more fine-grained sub-tasks, offering more optimization space; 3) a DAG runtime executor can implement RAD withour requiring the consistent low-level ML framework versions. To enhance system efficiency, we implement a workload estimator and design an OP-Fence scheduler to cluster devices with similar bandwidths together and partition the DAG to increase throughput. Additionally, we propose an AdaTopK compressor to adaptively compress intermediate activations and gradients at the slowest communication links. To evaluate the convergence and efficiency of our system and algorithms, we train ResNet-101 and GPT-2 on three real-world testbeds using 48 GPUs connected with 8 Mbps~10 Gbps networks. Experimental results demonstrate that our system and method can achieve 1.45 - 9.39x speedup compared to baseline methods while ensuring convergence.
format Preprint
id arxiv_https___arxiv_org_abs_2410_12707
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression
Tang, Zhenheng
Kang, Xueze
Yin, Yiming
Pan, Xinglin
Wang, Yuxin
He, Xin
Wang, Qiang
Zeng, Rongfei
Zhao, Kaiyong
Shi, Shaohuai
Zhou, Amelie Chi
Li, Bo
He, Bingsheng
Chu, Xiaowen
Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
To alleviate hardware scarcity in training large deep neural networks (DNNs), particularly large language models (LLMs), we present FusionLLM, a decentralized training system designed and implemented for training DNNs using geo-distributed GPUs across different computing clusters or individual devices. Decentralized training faces significant challenges regarding system design and efficiency, including: 1) the need for remote automatic differentiation (RAD), 2) support for flexible model definitions and heterogeneous software, 3) heterogeneous hardware leading to low resource utilization or the straggler problem, and 4) slow network communication. To address these challenges, in the system design, we represent the model as a directed acyclic graph of operators (OP-DAG). Each node in the DAG represents the operator in the DNNs, while the edge represents the data dependency between operators. Based on this design, 1) users are allowed to customize any DNN without caring low-level operator implementation; 2) we enable the task scheduling with the more fine-grained sub-tasks, offering more optimization space; 3) a DAG runtime executor can implement RAD withour requiring the consistent low-level ML framework versions. To enhance system efficiency, we implement a workload estimator and design an OP-Fence scheduler to cluster devices with similar bandwidths together and partition the DAG to increase throughput. Additionally, we propose an AdaTopK compressor to adaptively compress intermediate activations and gradients at the slowest communication links. To evaluate the convergence and efficiency of our system and algorithms, we train ResNet-101 and GPT-2 on three real-world testbeds using 48 GPUs connected with 8 Mbps~10 Gbps networks. Experimental results demonstrate that our system and method can achieve 1.45 - 9.39x speedup compared to baseline methods while ensuring convergence.
title FusionLLM: A Decentralized LLM Training System on Geo-distributed GPUs with Adaptive Compression
topic Distributed, Parallel, and Cluster Computing
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2410.12707