bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders

Deep learning models involve computationally intensive training experiments. Increasing the batch size improves the training speed and hardware efficiency by enabling deep neural networks to ingest and process more data in parallel. Inspired by learning rate adaptation policies that yield good resul...

Full description

Saved in:
Bibliographic Details
Main Authors: George Stoica, Mihaela Elena Breabăn
Format: Article
Language:English
Published: Elsevier 2025-05-01
Series:SoftwareX
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2352711025001293
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849718816393658368
author George Stoica
Mihaela Elena Breabăn
author_facet George Stoica
Mihaela Elena Breabăn
author_sort George Stoica
collection DOAJ
description Deep learning models involve computationally intensive training experiments. Increasing the batch size improves the training speed and hardware efficiency by enabling deep neural networks to ingest and process more data in parallel. Inspired by learning rate adaptation policies that yield good results, methods that gradually adjust the batch size have been developed. These methods enhance hardware efficiency without compromising generalization performance. Despite their potential, such methods have not gained widespread popularity or adoption: unlike widely used learning rate policies, for which there is built-in support in most of the deep learning frameworks, the use of batch size adaptation policies requires custom implementations. We introduce an open-source package that implements batch size adaptation policies, which can be seamlessly integrated into deep learning training pipelines. This facilitates more efficient experimentation and accelerates research workflows.
format Article
id doaj-art-a312605764ca4e9c8234c7d8316a8283
institution DOAJ
issn 2352-7110
language English
publishDate 2025-05-01
publisher Elsevier
record_format Article
series SoftwareX
spelling doaj-art-a312605764ca4e9c8234c7d8316a82832025-08-20T03:12:16ZengElsevierSoftwareX2352-71102025-05-013010216210.1016/j.softx.2025.102162bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoadersGeorge Stoica0Mihaela Elena Breabăn1Corresponding author.; Alexandru Ioan Cuza University, Faculty of Computer Science, Iasi, RomaniaAlexandru Ioan Cuza University, Faculty of Computer Science, Iasi, RomaniaDeep learning models involve computationally intensive training experiments. Increasing the batch size improves the training speed and hardware efficiency by enabling deep neural networks to ingest and process more data in parallel. Inspired by learning rate adaptation policies that yield good results, methods that gradually adjust the batch size have been developed. These methods enhance hardware efficiency without compromising generalization performance. Despite their potential, such methods have not gained widespread popularity or adoption: unlike widely used learning rate policies, for which there is built-in support in most of the deep learning frameworks, the use of batch size adaptation policies requires custom implementations. We introduce an open-source package that implements batch size adaptation policies, which can be seamlessly integrated into deep learning training pipelines. This facilitates more efficient experimentation and accelerates research workflows.http://www.sciencedirect.com/science/article/pii/S2352711025001293Deep learningLearning rateBatch sizeSchedulers
spellingShingle George Stoica
Mihaela Elena Breabăn
bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
SoftwareX
Deep learning
Learning rate
Batch size
Schedulers
title bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
title_full bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
title_fullStr bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
title_full_unstemmed bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
title_short bs-scheduler: A Batch Size Scheduler library compatible with PyTorch DataLoaders
title_sort bs scheduler a batch size scheduler library compatible with pytorch dataloaders
topic Deep learning
Learning rate
Batch size
Schedulers
url http://www.sciencedirect.com/science/article/pii/S2352711025001293
work_keys_str_mv AT georgestoica bsschedulerabatchsizeschedulerlibrarycompatiblewithpytorchdataloaders
AT mihaelaelenabreaban bsschedulerabatchsizeschedulerlibrarycompatiblewithpytorchdataloaders