Adjustments for training with different models

Some adjustments. All tests pass and we have a new way to deal with different loss functions for different models. Validation is working but will open new Issue for further improvements.