Saving parameters of a Bayesian model (posterior) and loading them as the Prior

Is there a way to save the parameters of a Bayesian neural network (each neuron has two parameters, mean and variance) as we save the parameters of a normal neural network by state_dict() ? is there any easier way than overwriting existing classes and modules of Pytorch?

No, pytorch is basically a framework for array computations, not a “probabilistic programming language” by itself. “distributions” module has a limited scope and won’t help you with priors, there is pyro project, more suitable for Bayesian models, though I don’t like its slowness.