Parametrization is done via Hydra.

Besides Hydra default logging mechanism, a custom logging decorator is implemented. Custom logger targets logging of train/test losses along with the configuration of a given run. Decorator logging is based on config.py dataclasses. Columns are created based on dataclass values and types. It combines selected model (DnCNNConfig for example) dataclass recursive parameter retrieval with LogConfig dataclass which contains train/test loss, epoch values.

I would suggest using litecli (as seen in requirements-dev.txt) to quickly look into the logged data.