Skip to main content
SHARE
Publication

MDLoader: A Hybrid Model-driven Data Loader for Distributed Deep Neural Networks Training

by Jonghyun Bae, Jong Youl Choi, Massimiliano Lupo Pasini, Kshitij V Mehta, Khaled Ibrahim
Publication Type
Conference Paper
Book Title
2024 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)
Publication Date
Page Numbers
1193 to 1195
Publisher Location
New Jersey, United States of America
Conference Name
2024 IEEE International Parallel and Distributed Processing Symposium Workshops (IPDPSW)
Conference Location
San Francisco, California, United States of America
Conference Sponsor
IEEE
Conference Date
-

In this work, we propose MD Loader, a hybrid in-memory data loader for distributed deep neural networks. MDLoader introduces a model-driven performance estimator to automatically switch between one-sided and collective communication at runtime.