The originial PyMC3 example also runs this model with auto-assigned NUTS for both parameters, for comparison.
In your script, when I modfiy step = [ConjugateStep(p, counts, tau)] to step = [], the sampler crashes during initialization with SamplingError: Initial evaluation of model at starting point failed!. Just before that, it issues multiple RuntimeWarning: invalid value encountered in log return np.log(x).
What needs to be adjusted in the script to make this run just as it did on the original PyMC3 example?