`__origin__` `AttributeError` when using `torch.library.custom_op`

jit
1

When I try to use torch.library.custom_op to decorate an existing function in the spconv library, I get an error AttributeError: __origin__.

The method I am decorating is an exact copy of the ops.get_indices_pairs method. I’ve added return typehinting and a metakernal.

My code looks like this:

@torch.library.custom_op("mylib::get_indice_pairs", mutates_args=())
def get_indice_pairs(...) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
   ...

@get_indice_pairs.register_fake
def _(...) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
  return torch.empty(), torch.empty(), torch.empty()

Can anyone shed any light on this? The fact that it is looking for __origin__ would seem to indicate to me that it is about a generics issue, but I have no idea how.