Location: Online
Eventbrite Registration: https://www.eventbrite.com.au/e/tensorflow-profiling-workshop-tickets-884669198387?aff=ebdssbdestsearch
About this event
Basic profiling and optimisation of TensorFlow / Keras tasks running on the national Machine Learning eResearch Platform by Oliver Cairncross (o.cairncross@uq.edu.au)
The Monash eResearch Centre in collaboration with the University of Queensland’s Research Computing Centre is conducting a hands-on profiling workshop for researchers working with TensorFlow or Keras on the national Machine Learning eResearch Platform (MLeRP). The workshop will cover basic code profiling and optimisation to make the best use of the platform.
This workshop is aimed for researchers that have little or no experience with profiling. Nevertheless, participants are expected to have general experience with Python, Jupypter Notebooks and machine learning. Attendees will gain a practical understanding of sizing tasks with respect to memory constraints and how to avoid bottlenecks that lead to poor performance.
The workshop will focus on simple techniques and will avoid advanced (or intermediate topics). Areas that require advanced expertise are highlighted but are not covered in depth.
The workshop is presented and promoted as part of an ML4AU Community of Practice series. MLeRP is a collaboration between Monash University, University of Queensland and QCIF. MLeRP received investment (https://doi.org/10.47486/NML01) from the Australian Research Data Commons (ARDC). The ARDC is funded by the National Collaborative Research Infrastructure Strategy (NCRIS).
Workshop Highlights
Understanding the environment, setup and general workflow
Profiling basics using TensorFlow’s built-in tools
Managing large profiling datasets
Identify bottlenecks to speed up run times
Configuration and sizing tasks for optimal use of resources
Simple optimisation strategies
Attendees
The workshop is open to Australian and New Zealand researchers. Researchers must have access to the MLeRP platform.
Basic Python programming ability is required.
Some experience interacting with JupyterLab environment.
Attendees will use their own workstation / laptop and participate in the workshop remotely using Zoom.