Hi,
I checked https://pytorch.org/docs/master/nn.html?highlight=layernorm#torch.nn.LayerNorm but it’s not very clear if normalized_input should be something like (x - E[x])/(V[X] + e) or normalized by dividing the last dimension by the euclidian norm or if we just have to give a tensor x without any normalization operation.
What is expected ?
Thank you !