Is there a way to register a custom container to be handled like a tensor/tensordict?

[mctigger]

Hi,
when extending PyTorch as shown here: https://docs.pytorch.org/docs/stable/notes/extending.html#extending-torch-python-api

How can I tell tensordict to handle my custom object similar to a tensor, e.g. applying torch.stack etc. correctly? It would be fine, if an exception is thrown for stuff that is not implemented.

Best regards!

[vmoens]

That is already the case, no? stack, cat etc are supported and you'll get a NotImplementedError if it's not.