I ran a pairs plot and it looks like all the response_mu parameters are highly correlated which is going to cause problems for the sampler. That makes my first thought that the model is over-parameterized, and can be expressed in a simpler way.
I looked up the a dose-response model in BDA3 (page 75) and it looks different than this model. I guess your responses are not integers but the probabilities between 0 and 1, which explains the beta distribution instead of the binomial. Is this how the data came or can you get number of deaths out of number of trials (integers)?
