Parallel sampling without pickle in version 4.0.0b3

Oh single core works fine. I initially switched to cores=1 to set this problem aside. But I’m starting to need the performance now.