Huawei's AI Research Unit has firmly rejected allegations that Pangue Pro Moe's leading language model copied elements from Alibaba's Qwen 2.5.
According to a report by News.az, the charges surfaced after an anonymous entity called Housentagi posted a report on Github, citing “extraordinary correlations” on Github.
The report argued that Huawei's model was not trained from scratch, but was “upcycled” from Alibaba's Qwen, raising concerns about possible copyright infringement and technical data forgery.
In response, Huawei's Noah Ark Lab issued a statement denying fraud. Pangue Pro Moe was developed independently, fully trained with Huawei's own Ascend Chips and said it was built with innovations in architecture and technical design. The lab also said it complies with all open source licensing requirements, but did not specify which model or codebase was referenced.
Alibaba did not respond to a request for comment and Reuters were unable to verify Honestagi's identity.
This argument comes amid growing competition in China's AI sector. While Alibaba's Qwen uses the Chatbot feature to target consumer use, Huawei's Pangu model focuses more on corporate and government applications. Huawei opened sourced Pangue Pro Moe in June, pushing for increased developer recruitment.
This controversy highlights the growing transparency and scrutiny of the AI model in increasingly crowded and competitive areas and on intellectual property.
news.az
