I added:
print(inputs)
print([type(a) for a in inputs])
print([a.dtype for a in inputs])
to perform, and when running the example above the output is:
Auto-assigning NUTS sampler...
Initializing NUTS using jitter+adapt_diag...
[array(1.), array(2.), array(2.08049756), array([5.46187651, 5.20988144, 4.94399615])]
[<class 'numpy.ndarray'>, <class 'numpy.ndarray'>, <class 'numpy.ndarray'>, <class 'numpy.ndarray'>]
[dtype('float64'), dtype('float64'), dtype('float64'), dtype('float64')]
Segmentation fault
The inputs are of the expected shapes and correspond to start, stop, a, and b.
Unless I am missing something, this looks as it should, so I guess the problem must be somewhere else (?).