Create a base-class for every hydra setup possibility

This base-class can be used for IDEs and type-checkers at initialization

from mdml_tools.hydra_models.optimizer import BaseOptimizer
import hydra

optimizer: BaseOptimizer = hydra.utils.initialize(cfg.optimizer)

This way the IDE knows that every next reference to optimizer belongs to the optimizer class.

the BaseOptimizer could be something like this (pseudo-code):

from pytorch_lightning.optimizers import BasicOptimizer

class BaseOptimizer(BasicOptimizer):
   pass

So the IDE knows its a torch optimizer, but when you follow the stack trace you will come back to where the hydra optimizers are created.