Checkout the script train.py. You will notice several blocks that check whether we want to perform fine tuning. If that is the case, the script will look for a registered model in the AML Model Registry, instead of training from scratch.
Note the additional input argument freeze_layers to the script. This allows us to explore which layers get to learn during fine-tuning.