Created by: pkuyym

First we compare the impact for larger language model under default golden settings.