As the later versions of v3 were being released, I struggled with the fact that pm.sample() was encouraging the use of InferenceData instead of the older MultiTrace object it had been returning. I started to gather notes about how to work with InferenceData and with the release of v4 I figured I would polish and share them. As is described in post, this should be seen as a complement to @OriolAbril 's post covering Arviz, PyMC, coords and dims. My notes are narrower and explicitly compare the older and newer ways of dealing with posterior samples. You can think of this as a bit of a transition guide narrowly addressing InferenceData object. Hope it helps someone!
Coordinates in PyMC & InferenceData Objects – Christian Luhmann – Python, math, etc.