Parallel Training of Pre-Trained Models via Chunk-Based Dynamic Memory Management
Author:
Affiliation:
1. Tencent Inc., Shenzhen, Guangdong, China
2. National Singapore University, Singapore
Publisher
Institute of Electrical and Electronics Engineers (IEEE)
Subject
Computational Theory and Mathematics,Hardware and Architecture,Signal Processing
Link
http://xplorestaging.ieee.org/ielx7/71/9956797/09940581.pdf?arnumber=9940581
Reference36 articles.
1. A short proof of optimality for the min cache replacement algorithm;roy;Inf Process Lett,2007
2. Evaluation techniques for storage hierarchies
3. A study of replacement algorithms for a virtual-storage computer
4. TurboTransformers
5. Colossal-AI: A unified deep learning system for large-scale parallel training;li,2021
Cited by 13 articles. 订阅此论文施引文献 订阅此论文施引文献，注册后可以免费订阅5篇论文的施引文献，订阅后可以查看论文全部施引文献
1. A Multidimensional Communication Scheduling Method for Hybrid Parallel DNN Training;IEEE Transactions on Parallel and Distributed Systems;2024-08
2. Training and Serving System of Foundation Models: A Comprehensive Survey;IEEE Open Journal of the Computer Society;2024
3. Hanayo: Harnessing Wave-like Pipeline Parallelism for Enhanced Large Model Training Efficiency;Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis;2023-11-11
4. Multi-level Storage Optimization for Intermediate Data in AI Model Training;Lecture Notes in Computer Science;2023-11-07
5. Towards GPU Memory Efficiency for Distributed Training at Scale;Proceedings of the 2023 ACM Symposium on Cloud Computing;2023-10-30
1.学者识别学者识别
2.学术分析学术分析
3.人才评估人才评估
"同舟云学术"是以全球学者为主线，采集、加工和组织学术论文而形成的新型学术文献查询和分析系统，可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容，当前同舟云学术共收录了国内外主流学术期刊6万余种，收集的期刊论文及会议论文总量共计约1.5亿篇，并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询！咨询电话：010-8811{复制后删除}0370
www.globalauthorid.com
TOP
Copyright © 2019-2024 北京同舟云网络信息技术有限公司 京公网安备11010802033243号 京ICP备18003416号-3