Hi @matiyanez
Similar to my comments in this other post, we leave the deployment part of trained models up to the user. If you want to stay in the Modulus workflow, Modulus provides some inference and validators you can use (Many of our examples use these, please have a look at the example documentation). Define these and add them to your training domain.
If you only have inferencers and validators added to your training domain, you can call traner._eval() (solver.eval()) to execute these without any training loop.