s).to(
        ...     torch.float
        ... )
        >>> loss = model(**inputs, labels=labels).loss
        >>> loss.backward()  # doctest: +IGNORE_RESULT
        ```NrĂ