The first approach we can use to save model parameters is using the torch::save function, which recursively saves parameters from the passed module:

 torch::save(model, "pytorch_net.pt"); 

To use it correctly with our custom modules, we need to register all the submodules in the parent one with the register_module module's method.

To load the saved parameters, we can use the torch::load function:

 Net model_loaded;
 torch::load(model_loaded, "pytorch_net.pt"); 

The function fills the passed module parameters with the values that are read from a file.