After AWS Autopilot creates a model, How to use that model to the training data set offline?
How to use that .tar.gz model file?
The .tar.gz file is a model artifact
To create a model, you combine the algorithm container and the model artifact
You can do so in the Console, under Inference > Models > Create Model
What do you mean by "How to use that model to the training data set offline?"
If you mean, "run a batch transformation", then once you create a model, you can select the model in the console, then click 'Create batch transform job'
If you want to do it locally, then you can use the SageMaker Python SDK in Local Mode and run the transform on your local computer (requires Docker)