PyTorch Lightning AUROC value for multi-class seems to be completely off compared to sklearn (using it wrong)?

Are you categorizing your predictions in the same way when you call the scikit-learn function?