Aboat remove_output in train.py

[Gwencong]

Why remove_output is set to True in the code train.py, it's strange to remove the trained checkpoints in output directory when training process is over.

[andreaconti]

Hi, this is cause I usually remove them and store only the final checkpoint as an artifact on mlflow to not clog the hard disk when multiple experiments are performed sequentially, but you are free to keep them if you like.