redesign to accomodate this use case

I have various python machine learning models/classes that have the following class defs:

class Model(pl.LightningModule):
   def __init__(self, **kwargs):
       # perform initialisation   

I am using pytorch lightning which also allows one to create a model instance from a saved checkpoint.

For each class I provide a load method as:

def load(**kwargs):
    if "checkpoint_file" in kwargs:
        # call the class method provided by the parent
        return Model.load_from_checkpoint(kwargs["checkpoint_file"])
    else:
        return MyModel(**kwargs)

I need to do this for every class as I need to provide a single callable instansiator (due to use of facebook hydra). I was wondering if there is some design pattern I can utilise to not have to define this for every class.