Playing around with this, I tried chol=np.zeros(1) and got ValueError: chol must be two dimensional. and the same if I used cov instead - which makes sense to me, because what does a single dimensional covariance matrix mean? sd*np.eye(D) is only giving a single value when D is 1 though. The ValueError: chol must be two dimensional should probably be thrown instead in your case - possibly a bug that it’s not, @pymc_devs_bot?
That said, I did get it to run (albeit with acceptance probability warnings) using sd = pm.HalfCauchy("sd", beta=0.1, shape=(1, 1)) - though I think that still just gives a single value, so probably I’m misunderstanding something.
10/10 for a perfect example of how to ask a question involving code, by the way.