Support for torch.nn.Parameter in nb::ndarray

[chrsmcgrr]

Hello,

I've been trying out the nb::ndarray feature with PyTorch and it works great. However, I noticed for torch parameters it does not seem to work with the ndarray interface.

Now parameters are a subclass of torch.Tensor and are used to hold the model state/weights.

I was wondering if it would be possible to extend the type check in ndarray. And if it would be possible to add a isinstance check? I understand the type check is done with strings to remove the dependency. But it would be great to have this feature as under the hood parameters will also have a DLPack implementation.

What does everyone think?