Gradient Weights for Softmax Function

[sidml]

Hi. In the paper you derived equations for the case when softmax function is used to obtain final class probabilities. (Equations 21-24)
I am unable to find an implementation of these formulae in this repository. Any help is greatly appreciated.