An option -bk 0 for turning off the backups of the model during training

[Twenkid]

Is your feature request related to a problem? Please describe.

I couldn't find an option for turning off saving backups (.bk) for faceswap.py train. The -ss is for snapshots, but I didn't find a way to turn off the .bk file. A reason for having that one is that IMO it is superfluous for educational experimenting and not important sessions, while it adds more writing to a disk or takes up space on a RAM disk - the effect is multiplied if one is trying out different models, settings, datasets and keeping their current states.

Describe the solution you'd like

An option, say:

-bk 0

...with default being 1.

Describe alternatives you've considered

I made a simple work around for myself by adding a flag in train/model/_base.py which is set in the code:

https://github.com/Twenkid/faceswap/blob/master/plugins/train/model/_base.py

DISABLE_BACKUP = True 
...
def _save(self):
        """ Backup and save the model and state file.
        
 if save_averages and self._should_backup(save_averages) and not DISABLE_BACKUP:
   ...

[torzdf]

Honestly, only 1 backup is kept, and the backup is only created when loss on both sides has dropped between save iterations to it's lowest average level. This is not a huge overhead on disk space or time.

I will tag this as "suggestion" but I cannot see myself implementing this any time soon. As always, I welcome pull requests