Huawei's Noa Ark Lab refutes claims of copying Alibaba's model

Beijing/Shanghai (Reuters) - Huawei's artificial intelligence research division, Noa Ark Lab, has denied allegations that a version of its PangU Pro large language model copied elements from Alibaba's model. The division stated that its model was independently developed and trained, without any reliance on external sources.

The statement comes in response to a paper posted on GitHub by an entity called Honestagi, which raised concerns about the similarities between Huawei's PangU Pro MOE model and Alibaba's Qwen 2.5 14b model. The paper suggested that Huawei's model may have been derived through 'upcycling', prompting discussions in AI circles and tech-focused media.

Noah Ark Lab emphasized that its model was not based on incremental training of other manufacturers' models and highlighted key innovations in architecture design and technical features. The lab's development team strictly adhered to open-source license requirements for any third-party code used.

Both Huawei and Alibaba are key players in the AI industry, with a focus on developing advanced language models. Huawei released its PangU Pro model in 2021, built entirely on Huawei's Ascend chips, while Alibaba's Qwen 2.5 14b model is part of its small-sized Qwen 2.5 family. The competition between these tech giants has spurred innovation and the release of open-source AI models, attracting developers and users worldwide.