attention问题

attention = Dense(1, activation='tanh')(char_embedding_dropout)
        attention = Flatten()(attention)
        attention = Activation('softmax')(attention)
        attention = RepeatVector(400)(attention)
        attention = Permute([2, 1])(attention)
        # apply the attention
        representation = multiply([BiLSTM, attention])
我想问一下这个attention是在lstm之前还是之后，第一次接触还请解答