For the refraction models in the Python package, I think the units should be hPa as reported in the paper and not Pa as shown in the documentation.
https://github.com/gschwind/sg2/blob/ca99a3ac35454ae85a4b2ec02c231c5bdebf2aa3/python/pysg2.cxx#L1046-L047
The Python functions are sg2.topocentric_correction_refraction_SAE and sg2.topocentric_correction_refraction_ZIM.
ICYMI @IoannisSifnaios and @kandersolar