Non-Centered Reparameterisation for the Negative Binomial Distribution?

Cool, thanks. Is there any best practice for setting the prior for dispersion?