Reports from some workloads indicate that the list form of the call is 4x faster than calling in a loop (when the loop involves recalling the model from cache like pgml).