There is random input to make the models evaluate differently every time, but without that (or if you fix it) then the model will always give the same output with the same input.

There's no somehow to it, it's actually surprisingly hard to make a computer do something in a non-deterministic fashion.