The attention weight calculation? #5
Description
It seems that the Attention layer is not properly computed. In the original paper, the vectors are computed as the weights sum of the weight and hidden state (h_i), but not the hidden representation (u_i).