Use soft floating point target label in nll_loss

vision
1

The nll_loss accept LongTensor as target is there any way to use FloatTensor as target. thanks in advance.