The loss of multi class

[Chenly1993]

Thank you for your contribution! Now ，I want to use it for 256 dimensional labels classification. Each label is 0 or 1. But the loss of training is strange. It will sharp drop at first. Then it will keep almost no changed.