Have a look at this post for a small example on multi label classification.
You could use multi-hot encoded targets, nn.BCE(WithLogits)Loss
and an output layer returning [batch_size, nb_classes]
(same as in multi-class classification).
11 Likes