Filed in the main repo here first, I run into a big endian dtype issue with Numba when computing forward models. A fix is suggested there