Qwen Technical Report

Qwen Technical Report

28 Sep 2023 | Jinze Bai, Shuai Bai, Yunfei Chu, Zeyu Cui, Kai Dang, Xiaodong Deng, Yang Fan, Wenbin Ge, Yu Han, Fei Huang, Binyuan Hui, Luo Ji, Mei Li, Junyang Lin, Runji Lin, Dayiheng Liu, Gao Liu, Chengqiang Lu, Keming Lu, Jianxin Ma, Rui Men, Xingzhang Ren, Xuancheng Ren, Chuanqi Tan, Sinan Tan, Jianhong Tu, Peng Wang, Shijie Wang, Wei Wang, Shengguang Wu, Benfeng Xu, Jin Xu, An Yang, Hao Yang, Jian Yang, Shusheng Yang, Yang Yao, Bowen Yu, Hongyi Yuan, Zheng Yuan, Jianwei Zhang, Xingxuan Zhang, Yichang Zhang, Zhenru Zhang, Chang Zhou, Jingren Zhou, Xiaohuan Zhou, Tianhang Zhu.
The Qwen Team from Alibaba Group introduces QWEN, a comprehensive series of large language models (LLMs) designed to revolutionize natural language processing tasks. QWEN includes base pretrained models and chat models finetuned with human alignment techniques, such as supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF). The base models demonstrate superior performance across various downstream tasks, while the chat models, particularly those trained with RLHF, are highly competitive. Specialized models for coding (CODE-QWEN) and mathematics (MATH-QWEN-CHAT) have also been developed, showing significant improvements over open-source models. The report details the pretraining, alignment, and specialized model development processes, along with experimental results and evaluations. QWEN's performance is compared against other LLMs, including proprietary models like GPT-4, and its capabilities in tool use, code interpretation, and agent applications are highlighted. The team emphasizes the importance of human evaluation and the need for new evaluation methods tailored to aligned models.The Qwen Team from Alibaba Group introduces QWEN, a comprehensive series of large language models (LLMs) designed to revolutionize natural language processing tasks. QWEN includes base pretrained models and chat models finetuned with human alignment techniques, such as supervised finetuning (SFT) and reinforcement learning from human feedback (RLHF). The base models demonstrate superior performance across various downstream tasks, while the chat models, particularly those trained with RLHF, are highly competitive. Specialized models for coding (CODE-QWEN) and mathematics (MATH-QWEN-CHAT) have also been developed, showing significant improvements over open-source models. The report details the pretraining, alignment, and specialized model development processes, along with experimental results and evaluations. QWEN's performance is compared against other LLMs, including proprietary models like GPT-4, and its capabilities in tool use, code interpretation, and agent applications are highlighted. The team emphasizes the importance of human evaluation and the need for new evaluation methods tailored to aligned models.
Reach us at info@study.space