How To Do Basic CI On Your ML Model With Sagemaker and Github-Actions

Here's a quick CI demo using Github-Actions and Amazon Sagemaker. While I'm using a toy example, with large data volumes and expensive compute instances, this is where Sagemaker really shines. It spins up my compute, trains and saves my model, and shuts everything down automatically.
Rather than run from Studio, Studio Lab, Canva, or the other myriad noteboook/IDE tools within the Sagemaker UI, I just use my local machine and my CI server to launch Sagemaker training jobs.
-- I develop and test locally, using the Sagemaker Python SDK.
-- On each PR, Github-Actions starts a CI job that builds my code and runs my basic model training step (also using the sg python sdk)
Next I'll layer on an explicit set of CD steps using different environments, as this another area where Sagemaker really helps.
This isn't to say that this process is painless - after all Continual needs to come in somewhere - but as an illustrative example of using a CI tool with Sagemaker (and not getting caught in the web of Sagemaker notebooks), I hope this helps