practical-pytorch

practical-pytorch copied to clipboard

The docs say that the embedding layer has the following input/output dimensionality: Input: LongTensor (N, W), N = mini-batch, W = number of indices to extract per mini-batch Output: (N, W, embedding_dim)

Yet, the tutorial gives as input for the embedding layer with the dimensionality W,N, and gets the dimensionality of the output as W,N, embedding_dim. If I understand this correctly (which I might not be), the order of the dimensions differs between your seq2seq batches example and the docs. If so, should you be transposing the matrices before you the inputs go through the embedding layer?