Adding custom attention layer in enocder-decoder architecture for neural machine translation using keras

Question

I am trying to do a neural machine translation task for converting english to hindi using tensorflow-2 Keras. I have created my own custom attention layer and i just cannot