Replicating a Generalized Linear Model fro unmixing

I would like to replicate a GLM for data with Poisson noise, whereby the deviance function is minimized (that assesses the goodness of fit by comparing he log-likelihood under the saturated model).