DesignCreator should inherit from torch.nn.Module

[julianhoever]

The abstract class DesignCreator should also inherit from torch.nn.Module because it is used to ensure that a torch.nn.Module always have a create_design function for translation. Therefore it is mostly used in combination with a torch.nn.Module.

elastic-ai.creator/elasticai/creator/vhdl/design_creator.py

Lines 6 to 8 in 662e45a

	class DesignCreator(ABC):
	@abstractmethod
	def create_design(self, name: str) -> Design: ...

In addition in the Sequential class we assume that submodules of type DesignCreator passed to the constructor are also torch.nn.Module, because we cast them to torch.nn.Module which is not possible if the submodules of type DesignCreator not inherit from torch.nn.Module.

elastic-ai.creator/elasticai/creator/nn/sequential/layer.py

Lines 12 to 27 in 662e45a

	class Sequential(DesignCreator, torch.nn.Sequential):
	def __init__(self, *submodules: DesignCreator):
	super().__init__(*cast(tuple[torch.nn.Module, ...], submodules))
	def create_design(self, name: str) -> Design:
	registry = _Registry()
	submodules: list[DesignCreator] = [
	cast(DesignCreator, m) for m in self.children()
	]
	for module in submodules:
	registry.register(module.__class__.__name__.lower(), module)
	subdesigns = list(registry.build_designs())
	return _SequentialDesign(
	sub_designs=subdesigns,
	name=name,
	)

Perhaps the name DesignCreator is not a good choice.

[glencoe]

I'm not sure i agree. For testing purposes and modularity, i like the two to be separated. But i agree that adding Module there would communicate intent better. What do you think about introducing a DesignCreatorModule or just CreatorModule (that users should inherit from) and keep the DesignCreator as is?