Huawei denies claims its AI model copied Alibaba's Qwen
Huawei's artificial intelligence research unit has firmly rejected allegations that its Pangu Pro Moe large language model copied elements from Alibaba’s Qwen 2.5.
The accusations surfaced after an anonymous entity called HonestAGI posted a report on GitHub, citing “extraordinary correlation” between the two models, News.Az reports, citing Reuters.
The report alleged that Huawei's model was not trained from scratch but instead "upcycled" from Alibaba's Qwen, raising concerns over copyright infringement and possible falsification of technical data.
In response, Huawei's Noah Ark Lab issued a statement denying any wrongdoing. It stated that Pangu Pro Moe was independently developed, fully trained on Huawei’s proprietary Ascend chips, and built with innovations in architecture and technical design. The lab also said it complied with all open-source license requirements, though did not specify which models or codebases were referenced.
Alibaba has not responded to requests for comment, and Reuters has been unable to verify the identity of HonestAGI.
The debate comes amid rising competition in China’s AI sector. While Alibaba’s Qwen targets consumer use with chatbot capabilities, Huawei’s Pangu models are more focused on enterprise and government applications. Huawei had open-sourced its Pangu Pro Moe in June in a push to increase developer adoption.
The controversy highlights the growing scrutiny over AI model transparency and intellectual property in an increasingly crowded and competitive field.
Similar news
