Lab 05: Troubleshooting & Testing (FSDL 2022)
In this video, we cover the fifth lab in the course, on how to test code and troubleshoot model performance issues. We focus on concrete workflows for testing and work through how to read the trace of a PyTorch training step and find easy-to-fix bottlenecks.
Try out the lab on Colab here:
https://fsdl.me/lab05-colab
Find all the labs on GitHub here:
https://github.com/full-stack-deep-le...
Subscribe to our channel or sign up at https://fullstackdeeplearning.com/cou... to follow along with the 2022 course!
00:00 Overview
00:51 Linting: pre-commit, black, flake8
05:42 Testing: pytest, doctest, memorization testing
11:15 Troubleshooting PyTorch performance
16:13 A guided tour of a PyTorch trace