Jointly train with two inputs

My data has [Nx2x10] shape and i want the model to take 2 inputs (xi and xj) at every time. when i stack the inputs xi and xj as (20,) the output probability is different from stacking xj and xi. How can i make sure that the probability is same for (xi,xj) and (xj,xi) ?

Can you share a code snippet on what you are trying to do?
It depends on your model and use case.

Probably you have to train the network with both the variants of your data, if you want the behaviour to be same. .