Deep Learning Frameworks

Help Desk

Video

This session gives an overview of using Tensorflow, Pytorch, and JAX are the core deep learning frameworks on ALCF production resources. All three frameworks are accessible in python (and a few other languages, for tf/torch) and offer the core elements of:

  • Automatic differentiation; 
  • GPU offload and acceleration from python;
  • Library of essential building blocks of machine learning operations;
  • Performant ways to scale codes out to multiple devices
  • An ecosystem of extensions and custom tools to make your life easier;
  • Export your trained models to open source inference engines (ONNX, etc)
  • All are open source; All will be supported on Aurora. Pick the one that makes sense for your problem!
Presenter
Corey Adams, ANL
Systems