Tried to use getattr(), but got NotImplementedError for some modules. Also, tried to use isinstance(instance, WeakScriptModuleProxy). Couldn't import it due to a recursive import. Any idea for this?

torch.jit depends on torch.nn since torch.jit.ScriptModule is a subclass of torch.nn.Module. I think if we want to support isinstance it has to be done only on WeakScriptModuleProxy (without making any changes to the inheritance chain of nn.Module). Maybe a less intrusive solution is to subclass ScriptMeta, add the __instancecheck__ method, and somehow install that on only the WeakScriptModuleProxy class

Agree, but isinstance() does not work as that way. When we do isinstance(module, nn.Linear), python does: type(nn.Linear).__instancecheck__(nn.Linear, module). That's why I added ModuleMeta to cover every module's __instancecheck__. We don't have an option implementing __instancecheck__ on ScriptMeta, refer to https://stackoverflow.com/questions/52168971/instancecheck-overwrite-shows-no-effect-what-am-i-doing-wrong for more information.
If we're not going this way, there's a method providing torch.isinstance() as I suggested.

Or... can we inherit from both class: SomeProxy(WeakScriptModuleProxy, original_cls), when creating WeakScriptModuleProxy instance?