AICB: A benchmark for evaluating the communication subsystem of LLM training clusters

AICB (Artificial Intelligence Communication Benchmark) is a benchmark for evaluating the communication subsystem of GPU clusters, which includes representative workloads in the fields of Large Language Model (LLM) training. Guided by the theories and methodologies of Evaluatology, we simplified the...

Full description

Saved in:
Bibliographic Details
Main Authors: Xinyue Li, Heyang Zhou, Qingxu Li, Sen Zhang, Gang Lu
Format: Article
Language:English
Published: KeAi Communications Co. Ltd. 2025-03-01
Series:BenchCouncil Transactions on Benchmarks, Standards and Evaluations
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2772485925000250
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849319608168742912
author Xinyue Li
Heyang Zhou
Qingxu Li
Sen Zhang
Gang Lu
author_facet Xinyue Li
Heyang Zhou
Qingxu Li
Sen Zhang
Gang Lu
author_sort Xinyue Li
collection DOAJ
description AICB (Artificial Intelligence Communication Benchmark) is a benchmark for evaluating the communication subsystem of GPU clusters, which includes representative workloads in the fields of Large Language Model (LLM) training. Guided by the theories and methodologies of Evaluatology, we simplified the real-workload LLM training systems through AICB that maintain good representativeness and usability. AICB bridges the gap between application benchmarks and microbenchmarks in the scope of LLM training. In addition, we constructed a new GPU-free evaluation system that helps researchers evaluate the communication system of the LLM training systems. To help the urgent demand on this evaluation subject, we open-source AICB and make it available at https://github.com/aliyun/aicb.
format Article
id doaj-art-4b3e69b5f7f44103bbda4403d00a0548
institution Kabale University
issn 2772-4859
language English
publishDate 2025-03-01
publisher KeAi Communications Co. Ltd.
record_format Article
series BenchCouncil Transactions on Benchmarks, Standards and Evaluations
spelling doaj-art-4b3e69b5f7f44103bbda4403d00a05482025-08-20T03:50:22ZengKeAi Communications Co. Ltd.BenchCouncil Transactions on Benchmarks, Standards and Evaluations2772-48592025-03-015110021210.1016/j.tbench.2025.100212AICB: A benchmark for evaluating the communication subsystem of LLM training clustersXinyue Li0Heyang Zhou1Qingxu Li2Sen Zhang3Gang Lu4Corresponding author.; Alibaba Cloud, Beijing, 100124, ChinaAlibaba Cloud, Beijing, 100124, ChinaAlibaba Cloud, Beijing, 100124, ChinaAlibaba Cloud, Beijing, 100124, ChinaAlibaba Cloud, Beijing, 100124, ChinaAICB (Artificial Intelligence Communication Benchmark) is a benchmark for evaluating the communication subsystem of GPU clusters, which includes representative workloads in the fields of Large Language Model (LLM) training. Guided by the theories and methodologies of Evaluatology, we simplified the real-workload LLM training systems through AICB that maintain good representativeness and usability. AICB bridges the gap between application benchmarks and microbenchmarks in the scope of LLM training. In addition, we constructed a new GPU-free evaluation system that helps researchers evaluate the communication system of the LLM training systems. To help the urgent demand on this evaluation subject, we open-source AICB and make it available at https://github.com/aliyun/aicb.http://www.sciencedirect.com/science/article/pii/S2772485925000250LLM training clusterBenchmarkCollective communicationDistribuited training
spellingShingle Xinyue Li
Heyang Zhou
Qingxu Li
Sen Zhang
Gang Lu
AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
BenchCouncil Transactions on Benchmarks, Standards and Evaluations
LLM training cluster
Benchmark
Collective communication
Distribuited training
title AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
title_full AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
title_fullStr AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
title_full_unstemmed AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
title_short AICB: A benchmark for evaluating the communication subsystem of LLM training clusters
title_sort aicb a benchmark for evaluating the communication subsystem of llm training clusters
topic LLM training cluster
Benchmark
Collective communication
Distribuited training
url http://www.sciencedirect.com/science/article/pii/S2772485925000250
work_keys_str_mv AT xinyueli aicbabenchmarkforevaluatingthecommunicationsubsystemofllmtrainingclusters
AT heyangzhou aicbabenchmarkforevaluatingthecommunicationsubsystemofllmtrainingclusters
AT qingxuli aicbabenchmarkforevaluatingthecommunicationsubsystemofllmtrainingclusters
AT senzhang aicbabenchmarkforevaluatingthecommunicationsubsystemofllmtrainingclusters
AT ganglu aicbabenchmarkforevaluatingthecommunicationsubsystemofllmtrainingclusters