torch.Tensor

Does it make sense to write Op @ x where x is a lazy linear operator and x a torch.Tensor ? Let me explain why I ask the question. I wrote a FWHT with a Pytorch, SciPy and lazylinop backend. In the case of backend='pytorch' I have to modify function like _sanitize_matmul(self, op, swap=False) to make the code works. Otherwize it returns raise ValueError('dimensions must agree') Indeed, op.size returns something like <built-in method size of Tensor object at 0x7f881044a4f0>. Do you think the best idea would be to cast x from torch.Tensor to np.ndarray before Op @ x ? Or to modify the code in such a way x could be a torch.Tensor ?