The InstrumentalVariableRegression.fit method overrides the ModelBuilder.fit method, but from what I can see this is unnecessary. And it actually is behind the curve because it doesn't grab the random_seed and provide it to pm.sample_prior_predictive or pm.sample_posterior_predictive.
Obviously check that local tests pass. As of right now we have some issues with doctests, see #323, and I'm actively working on that.
cc @NathanielF