TorchScript = (trained) model export or application export?

Hello World,

after excessively reading the ‘Loading a PyTorch model in c++’ I have some questions left. Is exporting the plain model the way to go, without training and testing it in Python? To me it seems like that what makes me kind of nervous when it comes to preprocessing the data with the functionalities given by numpy, which are actuallly really helpfull. Is there a way to export the key functionality to C++ transforming the code to TorchScript, including the preprocessing procedures? What ist the best practice here? I am not that familiar with C++ and would like to have as less export trouble as possible.

Thanks for reading und hopefully answering:)