My solution to put gradient boosting models into production in SAS is to use the gbm2sas function in the mlmeta R package. This function creates pure SAS code that runs without any dependencies on Python, and it does not even require SAS Enterprise Miner, SAS/IML, etc.    It does not use LightGBM, Catboost, or XGBoost. It uses GBM. I recently did a small test here comparing GBM vs. LightBoost. LightGBM was much faster, but GBM made more accurate predictions. You can also speed up GBM in SAS using various tricks such as model tuning and parallel processing (yes, in SAS). ... View more