Why the performance of pyspark script doesn't improve with more number of cores and executors?

Question

I have a script for binary classification by loading a pre-trained model. I wonder why when I try different combinations of num-executors and executor-cores, I always get ap