Benchmarking Python Deep Learning Frameworks for Language Modeling on GPUs

7 pages•Published: October 17, 2024

Abstract

Neuralnetworksareomnipresentinnaturallanguageprocessing(NLP). We benchmark three popular Python frameworks (DyNet, TensorFlow, and Theano) on the standard NLP task of language modeling, and find that DyNet is significantly faster on this task. We also discuss other bottlenecks beyond performance, such as ease of use, that may impact the selection of a neural network framework.

Keyphrases: benchmarking, deep learning frameworks, natural language processing (nlp), neural networks, performance metrics

In: Lindsay Quarrie (editor). Proceedings of 2024 Concurrent Processes Architectures and Embedded Systems Hybrid Virtual Conference, vol 20, pages 74-80.

Links:	https://easychair.org/publications/paper/jkgx
	https://doi.org/10.29007/ptjm

BibTeX entry

@inproceedings{COPA2024:Benchmarking_Python_Deep_Learning,
  author    = {Vijaya Laxmi Pachva},
  title     = {Benchmarking Python Deep Learning Frameworks for Language Modeling on GPUs},
  booktitle = {Proceedings of 2024 Concurrent Processes Architectures and Embedded Systems Hybrid Virtual Conference},
  editor    = {Lindsay Quarrie},
  series    = {Kalpa Publications in Computing},
  volume    = {20},
  publisher = {EasyChair},
  bibsource = {EasyChair, https://easychair.org},
  issn      = {2515-1762},
  url       = {/publications/paper/jkgx},
  doi       = {10.29007/ptjm},
  pages     = {74-80},
  year      = {2024}}

Download PDF Open PDF in browser