Add an attention layer to the top of the lstm modl

Question

I have a simple network of one LSTM and one dense layer. I have two input and 18 outputs. The code and model is as follows:

import time
import tensorflow as t