The STAN user guide includes an excellent discussion on reparameterization, including centered/non-centered. This blog post by Michael Betancourt might also interest you, it touches on reparameterization in a broader discussion of model identification, sampler degeneracy, and what to do when NUTS is going slow or diverging.
For PyMC specific, the case study on multilevel modeling example notebook has examples of centered and non-centered implementations for the univariate and multivariate cases. I got a lot out of this notebook, but it’s a bit hidden!