The development of MOSS includes three stages. (IMAGE)
Caption
In stage 1, researchers pre-train the cross-lingual MOSS-base model with public text and code corpora. In stage 2, they first perform supervised fine-tuning (SFT) with synthetic conversational data and deploy it to the public. They then use the collected real-world data as a seed set to synthesize a new training set, which is used to perform the final SFT. In stage 3, They train a preference model and use it to perform preference-aware training. The models resulting from the three stages are named MOSS-base, MOSS-SFT, and MOSS-PAT, respectively.
Credit
Beijing Zhongke Journal Publising Co. Ltd.
Usage Restrictions
Credit must be given to the creator.
License
CC BY