When you run learn.export, it only saves the model weights and some additional info required for inference (activation function, train and val augmentations, etc).
It is meant to store awak models for inference specifically.
If you want to train further with the exported model, I’d suggest something like:
learn_exp = load_learner(...) # load exported model
learn_train = cnn_learner(...) # create a new `Learner` with the same setup as the previous one
# Both of these should work. The first is safer
learn_train.model.load_state_dict(learn_exp.model.state_dict())
learn_train.model = learn_exp.model
In the future, if you want to load models and continue saving, it’s better to save the model rather than export it.