Neelay Shah

Error in saving "best" models based on running lowest validation metric / loss.

- Config and registry system - Training pipeline - Distributed training - Mixed precision and others.

Extend the CD workflow such that the version string variables for the package are automatically updated every time a release is created on GitHub. It should extract the version number...

Clean up requirements.txt

1

Modifies `requirements.txt` to only contain packages which are used in the library. Achieved using [pipreqs](https://github.com/bndr/pipreqs).

This would allow for the best metric and loss values to be carried over when resuming training.