Simplest way to train constant to use in R tfprobability distribution lambda layer

Question

In R keras and with tfprobability, I want to find a simpler way to train constants for parameters in probability distributions. As it is, I can set up a model with two input