Hey, thanks a lot for pointing this out. In pytorch I found it difficult to recursively call and do backpropagation as in the original implementation. So I did it in a different way.
LayerNorm has been recently released and I will add it in the upcoming updates.
recursive_generator
similar with the official code?LayerNorm
is in here http://pytorch.org/docs/master/nn.html#layernorm.(Seems in this week v0.4 will be released)