Negative binomial returning -inf logp

Questions
4

I was able to reduce the data to a single point which would give a “SamplingError: Bad initial energy” error. Strangely it was not the largest data point in the data set.

Y_OBS = 100                             # No error
Y_OBS = 561270;                      # Error
Y_OBS = 561270 - 1000;           # Error
Y_OBS = 561270 - 10000;         # No error
Y_OBS = 798568;                      # No error

mu = pm.Normal("mu", mu=0, sd=10)
Y = pm.NegativeBinomial("Y", mu=2**mu, alpha=10, observed=Y_OBS);

Any idea of why this happens?

Thanks!