There is also this example: Refitting PyMC3 models with ArviZ (and xarray) — ArviZ dev documentation which is a bit more manual and might help clarify what is going on to obtain the pointwise log likelihoood values. I am not sure into which issues you ran, it is probably a bit easier than it was, but it is still an open issue, see for example Do not collapse timeseries logp across time dimension · Issue #5741 · pymc-devs/pymc · GitHub