I can confirm that it does work for NLP though. I used exactly this syntax on my text classification data: learn = text_classifier_learner(data_clas, arch=AWD_LSTM, drop_mult=0.5, metrics=[accuracy, AUROC()])
I was using it for multi-label classification in vision task where it always raised an error. I’m not sure what led it into an error if it worked in text classification.
It has been about a month but I will just drop this here. AUROC in the fastai library is restricted to binary classification problems. Use it for multi-classification must be the cause of the error.