At the moment you can save the torch models trained with
qnn.NeuralNetRegressor by using
torch.save on the
module_ member of these classes. You can then re-instantiate a NeuralNetClassifier or NeuralNetRegressor based on this already trained classifier, by passing it to the constructor.
For the other scikit-learn compatible models you can not at the moment save the models. In the future we will implement FHE engine serialization so that you can save the compiled FHE circuit and re-use it in deployment.