PyTorch LogSoftmax vs Softmax for CrossEntropyLoss

Question

I understand that PyTorch\'s LogSoftmax function is basically just a more numerically stable way to compute Log(Softmax(x)). Softmax lets you convert the output