The current way of specifying inputs is to use a list of tensor names ``` in = [a, b, c] a { 1 } ... ``` It would be more consistent if there was a dedicated input tensor ``` input { a = 1, b = 2, c = 3, } ```