What should be the structure sister networks can share weights in Siamese network? (Tensorflow or Keras)

Question

So if I have to build a Siamese network in Keras, I can give 2 inputs to the model but the real problem is that when I give 2 separate inputs, the flow of gradients will be