Spot Challenge for CodeNet and High Performance Deep Learning

December 15, 2021 10:00 am CST

Join us to hear a reminder about how to get started with the Code Similarity Challenge followed by an advanced talk from Dr Ulrich Finkler Called Rings and Things. More details follow:

Part 1 – Get started with the Code Similarity Challenge – We’ll go over the getting started steps –

Starter Steps
Listen to:
Call for Code Spot Challenge for CodeNet – Introduction 2 – https://www.crowdcast.io/e/call-for-code-spot-9 
Call for Code Spot Challenge for CodeNet – Setting up your system – https://www.crowdcast.io/e/call-for-spot-challenge
Run https://github.com/IBM/Project_CodeNet/tree/main/Contest/ExampleSimAnalysis


Part 2 – High Performance Deep Learning : Rings and Things:
Ulrich will briefly introduce model and data parallelism, and then go through the core collective function (all-reduce in NCCL) that
is used and explain how it works under the covers. Ulrich will then go through the ‘detective story’ on how to get from a benchmark (synthetic imagenet) to a real image processing application looking at flexibility of use and how to get performance,
again going under the covers down to lock-free programming in a parallel data loading setup inside a python extension.

Note this advanced talk is not related to the code similarity challenge but we thought that some of the community would be interested to hear about this topic.

 

Speakers:

Ulrich Finkler, IBM Research

 

Legend