After discovering a bug in #20, I think that the instructor deserves some further cleanup especially regarding unused initialization (e.g. optimizer not needed if we don't want to train the model).
This requires checking every usage of the instructor object and making sure that we use a common api (e.g. not using the underlying model attributes if those are propagated by the instructor.
After discovering a bug in #20, I think that the instructor deserves some further cleanup especially regarding unused initialization (e.g. optimizer not needed if we don't want to train the model). This requires checking every usage of the instructor object and making sure that we use a common api (e.g. not using the underlying model attributes if those are propagated by the instructor.