First of all you should train your teacher and save it in a dict with this structure
state_dict ={"model_state_dict":{"generator_ab":# state dict here"generator_ba":# state dict here"discriminator_a":# state dict here"discriminator_b":# state dict here}torch.save("logdir/last.pth", state_dict)
After that the API is very similar to training API.
But there are several callbacks you should add. Let's take a look on this part of the pipeline.