Output of AdaptiveSoftmaxWithLoss

I need help regarding torch.nn.AdaptiveSoftmaxWithLoss
It returns both output and loss.
output is the computed target log probabilities for each example, but I can’t seem to make sense of this.
If it was a probability distribution over C classes then I could have predicted a particular class.
How do I interpret this single target log probability? And what can I do with this?