The Huging Face and machine learning platform introduced an updated rating the best open language models in the world. According to this rating, Chinese models now dominate in the overall standings, actively used in various industries, including consumer electronics, cars and online games.
The first place in the TOP-10 was taken by the QWEN-72B-Innstruct model from Alibaba Group Holding with an average score of 43.02 for six indicators. This model, trained at 72 billion parameters, is currently a leader among open models, as the co -founder and general director of the Huging Face Clement Clement.
The second place went to the LLAMA-3-70B-INSTRUCT AI-Model from Meta*. The third place was also taken by the model from Alibaba – QWEN -72B. Microsoft Copilot model, the main partner of Openai, is located in the sixth place.
The seventh place in the ranking was taken by the YI-1.-1.5-34B-Chat model from the Beijing Startup 01.Ai, founded in 2023 by the famous venture capitalist Lee Kaifu, who previously held the post of President of Google China. Closes the top ten another model from Alibaba – QWEN1.5-110B.
Thus, three of the four Chinese LLMs that fell into the top 10 were from the Tongyi Qianwen series (also known as QWEN) developed by Alibaba. This demonstrates the rapid progress of the company in the field of AI, largely due to its adherence to open development. Open source code provides public access to the program code, which gives developers the opportunity to change or share its design, eliminate problems or expand functionality.
Technical Director of Alibaba Cloud Zhou Jinzhen said in May that the family of large language models Tongyi Qianwen is now used by more than 90 thousand corporate clients.
It is worth noting that the methodology for ranking of AI models from Huging Face now takes into account more metrics, including solving complex problems with a volume of about 1000 words. However, since the rating is focused on open source models, it excludes advanced LLM, developed in a closed environment, such as Openai GPT.
For comparison with closed AI models, you can contact third-party shooting lines. So, in the ranking of LMSYS CHATBOT Arena, the GPT-4O leads, the last LLM from Openai, presented in May. In the same rating, the QWEN2-72B-In-In-In-In-In-In-In-In-In-In-In-In-Inpel from Alibaba took only 19th place.