Getting varying results on running Prophet on different times of the day on Databricks

Hi,
I am using parallel processing to run Prophet model on a set of combination of variables with one prophet model fit on one type of variable combination on Databricks using a spark dataframe for faster computation.

For eg, if I have 500 different types of combinations, I run 500 prophet models each fit on a different type of combination. Each model is fit and predictions are made parallelly for multiple prophet models at time and at the end all the outputs from all the models are collated to give one output.
But what I have noticed is when I run the same code, let's say once in the morning and once at night, I get quite different predictions on both the occasions. Same happens on running the code on different days as well.
Can someone please explain any reason behind this?