Created by: wojtuss

With this change, the default behavior of the accuracy benchmarking scripts for QAT image classification and QAT NLP will be to use the whole dataset instead of a single iteration.

test=develop