Hi, I have a question that I feel is not addressed in the Huggingface SageMaker documentation, if you look at the example code snippets from the models page to train a model in SageMaker you get stuff like this:
huggingface_estimator = HuggingFace( entry_point='run_glue.py', source_dir='./examples/pytorch/text-classification', instance_type='ml.p3.2xlarge', instance_count=1, ...
This leads me to understand that I am going to need at least two SageMaker instances, one to run this notebook and another to run this code inside sagemaker?
Does the “Driver” instance will also be running while it waits for the training job?
Should I use the cheapest possible instance as a driver? Can I run the training script “locally” using this snippet so that no new instances are launched?