Issues parallelizing pymc3 model with the `multiprocessing` library

As an addendum to the above, in lieu of the multiprocessing method, are there any recommended approaches to fitting pymc3 models in parallel on an N-cpu system?