PyTorch implementation of L-Softmax loss.
This quick and dirty implementation is getting some stars. Please note that I've not heavily tested the code, so if you want to use this in your research, business, or whatever, please re-check the correctness of the implementation.