Got different predicted values

Hello, thank you for making such a nice package.
I got different results when running the models and would like to understand the underlying reason and how to set up to get the results that are supposed to be.
1. When running the model on macOS and Windows, why I got a different predicted value? I already check the version of prophet and stan and the versions are the same.
2. When I use Prophet with version 0.6, I also got a different result from the current version 0.7?
3. When I apply add_regressors of more than 2 inputs to the model in a different order (e.g., Run1: add X1, X2, and then X3. Run2: add X2, X3, X1), why the predicted values are different between the Runs? Is the order of adding regressors matter?