Alibaba’s Tongyi Qianwen family secures three of the top four spots among Chinese AI models, highlighting the company’s advancements in generative AI technology.
Alibaba Group Holding’s large language models (LLMs), essential for generative artificial intelligence (AI) services like ChatGPT, have achieved significant global recognition. According to the latest rankings from AI and machine-learning developer platform Hugging Face, Alibaba’s Tongyi Qianwen series, also known as Qwen, boasts three of the top four positions among Chinese LLMs.
Hugging Face’s updated leaderboard, released on Wednesday, places Alibaba’s Qwen-72B-Instruct model at the very top. This model, trained on 72 billion parameters, achieved an impressive average score of 43.02 across six benchmarks. Parameters in machine learning refer to variables used during training that determine how data inputs generate the desired outputs.
“Qwen 72B [Instruct] is the king, and Chinese open models are dominating overall,” remarked Clement Delangue, co-founder and CEO of Hugging Face, in a post on X (formerly Twitter).
The Qwen-72B-Instruct model was particularly noted for its excellence in mathematics, long-range reasoning, and knowledge application, as highlighted by the Hugging Face developer community.
In addition to the top-ranked Qwen-72B-Instruct, Alibaba’s Qwen-72B and Qwen1.5-110B models were ranked third and tenth, respectively. The rankings on Hugging Face are subject to change as new models are evaluated.
The Yi-1.5-34B-Chat model from Beijing-based start-up 01.AI, founded in 2023 by prominent venture capitalist Lee Kai-fu, secured the seventh spot. Lee was formerly the president of Google China.
This recognition underscores Alibaba’s rapid progress and commitment to AI and open-source development. Over the past few months, Tongyi Qianwen has been made available to third-party developers, reflecting Alibaba’s support for the open-source community. Open-source technology, which allows public access and modification of a program’s source code, has significantly contributed to China’s tech industry growth.
Meta Platforms, the parent company of Facebook, secured the second spot on Hugging Face’s list with its AI model. Microsoft, a major backer of ChatGPT creator OpenAI, had one model ranked sixth.
Hugging Face’s updated methodology for ranking AI models includes more metrics, such as evaluating performance on complex problems approximately 1,000 words in length. This change comes as AI performance begins to plateau, necessitating more nuanced evaluation criteria.
It’s important to note that the Hugging Face ranking focuses on open-source models, excluding advanced LLMs developed in proprietary environments, such as OpenAI’s GPT.
Separately, the LMSYS Chatbot Arena ranking by the University of California, Berkeley, and the Large Model Systems Organisation, recently crowned OpenAI’s latest LLM, GPT-4, as the top model introduced in May. Alibaba’s Qwen2-72B-Instruct was ranked 19th in this list.
These achievements highlight the competitive landscape of AI development and Alibaba’s leading role in advancing open-source AI technologies.
Related topics:
OpenAI and Microsoft Face New Copyright Infringement Lawsuit from Investigative News Nonprofit
OpenAI and Time Announce Groundbreaking AI Training Partnership
ChatGPT-4 passes the “Turing Test” Scientists: AI intelligence is comparable to humans