Huawei's AI Research Division claims that the version of the Pangu Pro Large Language model was developed and trained independently, copying elements of the Alibaba model.
A division, known as Noah Ark Lab, issued a statement on Saturday. This issued a statement the day after an entity called Houshestagi submitted an English paper to code sharing platform Github, which stated that Houshawei's Pangue Pro Moe (a mixed expert) model had “extraordinary correlation” with Alibaba's Qwen 2.5 14b.
This suggests that Huawei's model is derived through “upcycling” and not trained from scratch, the paper says, prompting extensive discussion in AI Circle Online and in China's technology-focused media.
The findings added that the findings point to false claims about potential copyright violations, the manufacture of information in technical reports, and Huawei's investment in training models.
In a statement, Noah Ark Lab said the model was “not based on incremental training of models from other manufacturers,” and “has made significant innovations in architectural design and technical features.”
It added that this is the first large-scale model fully built on Huawei's ascend chip.
It also said that its development team was strictly adhering to open source licensing requirements for the third-party code used.
Alibaba did not respond immediately Reuters Request a comment. Reuters could not contact Honesty or learn who was behind the entity.
The release of Chinese startup Deepseek's open source model R1 in January this year shocked Silicon Valley at a low cost, causing fierce competition amongst China's tech giants to deliver competitive products.
The QWEN 2.5-14B was released in May 2024 and is one of Alibaba's small Qwen 2.5 model families that can be deployed on PCs and smartphones.
Huawei entered the massive language model arena early in 2021 with the original Pangue release, but has since been perceived as lagging behind its rivals.
In late June, they opened sourced the Pangue Pro MoE model on Chinese developer platform Gitcode, and tried to boost the adoption of AI technology by providing free access to developers.
Qwen is for consumers and has a chatbot service like ChatGpt, but Huawei's Pangu model tends to be used more in the government and financial and manufacturing sectors.
