Making predictions with trained model

Physics Simulation Modulus Physics-ML Model Framework Technical Support (Modulus Only)
1

Greetings,

I have had the same issue as described here Deploying/using a trained model | Post-Training.

A trained AFNO model is created following the this tutorial Darcy Flow with Fourier Neural Operator - NVIDIA Docs in combination with the rest of the examples of the page. My goal is to export the predictions of the test dataset in e.g. numpy format to be saved and post-processed. As stated in the link, we got 2 options to make predictions:

  1. Using built-in methods. The tutorials merge config, model and constraints in the Solver. How can we use the Solver built-in functions modulus-sym/modulus/sym/solver/solver.py at main · NVIDIA/modulus-sym · GitHub to make predictions apart from eval()? Eval() indeed produces visual output using the plotter, but I would also want to somehow get the raw output.
  2. Second option is to treat it like a PyTOrch model. Do we do this through the Solver? Could you please elaborate more on that?

Thanks a lot!