Built upon LlaMA-13b, this version incorporates pretraining weights from a secondary full-scale pretraining phase using both Chinese and English bilingual data. This augmentation improves the model's comprehension of Chinese. For further details, please refer to this link.
评论