ALCF AI Testbed: Graphcore AI Training Workshop

America/Chicago
Murali Emani (LCF) , Venkatram Vishwanath (LCF)
Description

This training workshop will introduce users to the novel AI accelerators deployed at the ALCF AI Testbed. This workshop is targeted at universities, national labs, and industries working on open science.

This training session will introduce users to the Graphcore Bow Pod system architecture and software stack. It will provide hands-on training to get started on the Bow Pod64 in the ALCF AI testbed. The event will be virtual and the sessions will be recorded and made available. 

NOTE: Please register and sign-up/reactivate your ALCF Account by 05/01 if you would like to participate in the hands-on session.  

Day 1 will focus on Bow Pod64 hardware and software architecture, Poplar and PopART; running models on Pod64 at ALCF AI Testbed.

Day 2 will focus on a deep dive into graph neural networks, large language models, profiling with PopVision, a hands-on session, and best practices. 

To Access ALCF machines please see the instructions below. Deadline to complete the following is 5/01:

  • New Users: Request an ALCF Computer User Account if you do not currently have one. Specify "aitestbed_training" as the project name.
  • Previous Users: If you have an ALCF Account that is currently inactive, submit an account reactivation request. Specify "aitestbed_training" as the project name.
  • Current Users: If you have an active ALCF account, click Join Project to submit a membership request for the project "aitestbed_training".
Registration
Participants
    • 1:00 PM 1:15 PM
      Introduction 15m
    • 1:15 PM 1:45 PM
      Graphcore BowPod64 Hardware 30m
    • 1:45 PM 2:30 PM
      Software Stack: TensorFlow, PyTorch, and Poplar 45m
    • 2:30 PM 2:45 PM
      Break 15m
    • 2:45 PM 3:15 PM
      Porting applications with Poplar 30m
    • 3:15 PM 4:00 PM
      How to use Bow Pod64@ ALCF 45m
  • Wednesday, 12 June
    • 1:00 PM 1:45 PM
      Deep Dive on Graph neural networks and Large Language Models 45m
    • 1:45 PM 2:15 PM
      Profiling with PopVision 30m
    • 2:15 PM 2:30 PM
      Break 15m
    • 2:30 PM 3:15 PM
      Hands-on session 45m
    • 3:15 PM 4:00 PM
      Best Practices, Q&A 45m