About "tf.reduce_mean(input_vectors, axis = 1)"

[ZhuYuHe]

when compute x = (1/N) sum_i xi， I notice that you use tf.reduce_mean(input_vectors, axis = 1).
However, N in the formula is the length of sentence, but tf.reduce_mean compute x use the length of sentence after padding.
So, the correct method to compute x should be sum(x)/len(x).