For executing the code, you can call it without any arguments. In this case, the code will train a model for each point of the assignment, and it will evaluate it. It will save the model as best_model.pt in the bin directory. If that file already exists, it will add a suffix to the name to avoid overwriting it.

Still, there are some optional arguments that you can use to customize the execution of the code. The optional arguments are:

• -d, --device: The device to be used for the various experiments (default: 'cuda:0')
• -s, --save: Save the best model (obtained during the training) to disk (default: False)
• -e, --eval_only: Evaluate only the model saved on the disk, without training any new model (default: False)
• -m, --model: The model path to be evaluated (used only with eval_only) (default: ./bin/best_model.pt)