Why is torch.cuda.OutOfMemoryError not a valid error class?

Question

I have the following code:

        try:
            # faster, but requires more memory
            G = self.sparse.to_dense().t() @ self.sparse.to_dense()
        except torch.cuda.OutOfMemoryError:
            # slower, but requires less memory
            G = torch.sparse.mm(self.sparse.t(), self.sparse)

My pylance seems to think that torch.cuda.OutOfMemoryError is not a valid error class. (See image.)

However, when I run the code, the torch.sparse.mm runs, showing that the exception was detected.

Why does pylance think that it's invalid when it clearly works?

Answer 1

The reason for this issue is that torch.cuda.OutOfMemoryError does not extend Python's Exception class. This issue has been fixed by PyTorch now. 109961