- The veteran AI researcher takes over Qwen after former technical chief Lin Junyang left in March
- Liu, an early Qwen core member, has led pretraining and helped develop several major model series
Alibaba Group has appointed veteran AI researcher Liu Dayiheng (刘大一恒) as head of its Qwen large language model project, according to two employees familiar with the matter, filling the role about six months after former technical chief Lin Junyang (林俊旸) left the company.
The appointment was also announced on Alibaba’s official Apsara Conference website, which lists Liu as head of the Qwen LLM project at Alibaba’s ATH business group and Token Foundry.
Liu also delivered a keynote on September 22 at the conference held in Hangzhou, directly after Alibaba CEO Eddie Wu.
Alibaba’s official conference agenda confirms the title and Liu’s new role.
Qwen gets a new leader
Lin, Qwen’s former technical head, left Alibaba in March to start his own venture.
Alibaba Cloud CTO Zhou Jingren (周靖人) subsequently oversaw the model team, while Liu continued to lead pretraining and also took charge of post-training and coding teams, according to Chinese media reports.
The changes came as Alibaba reorganized its AI businesses. In June, Alibaba merged its Qwen-related business with its Future Life Lab to form the Token Foundry unit, headed by Wu.
From Qwen’s early days
Liu has been a core member of the Qwen team since its early development.
He earned his computer science doctorate from Sichuan University and joined Alibab’s DAMO Academy in 2021 after being selected for Huawei’s “Genius Youth” program in 2020.
He has been involved in the development of Qwen’s core language models as well as Qwen-Coder and multimodal models.
His appointment gives Qwen a leader who has been closely involved in the model family’s development rather than an executive brought in from outside the team.
At the Apsara Conference, Liu’s keynote was titled “Qwen: Toward Real-World Agents,” highlighting Alibaba’s push to extend Qwen from model capabilities toward AI agents that can perform tasks and interact with real-world applications.
Header image courtesy of “AI Lanmeihui“
