I'm trying to understand if the model performs equally well when prompting in English and in Chinese, is there data on this?