Yes unfortunately the number of parameters is pretty high in my model and this is related to a thread that I opened earlier. I looks like it is not possible to enforce Pyro to estimate a posterior only for a subset of the variables in the sample site and I have to use MCMC methods unfortunately … .