Why do you use the nnl_loss?

[swagshaw]

Hi bro, no offense, I see you use the nnl_loss as the loss function, did you know the exact loss function in the paper?
And why do you use the nnl_loss here, not CE Loss and others?

Thank you very much.

[yangjiabupt]

Actually, the only difference between nnl_loss and ce_loss is that the input feeding nnl_loss have done log_softmax operation before