Huawei denies Pangu model was based on Alibaba's Qwen
Briefly

Huawei's new large language model, Pangu Pro MoE 72B, was released open-source in June and faced allegations from HonestAGI claiming it copied from Alibaba's Qwen 2.5 model. HonestAGI's research, published on GitHub, pointed to high similarities between the two models using fingerprinting and statistical analysis. Concerns arose regarding Huawei's training process and compliance with open-source license rules. Huawei's research arm defended the model's independent creation, citing innovations in architecture design and adherence to licensing, although details on input from open-source references were not disclosed. Scrutiny over AI model origins is intensifying with the growth of open-source AI.
Huawei's Pangu Pro MoE 72B model, released under an open-source license, faced allegations of copying parts from Alibaba's Qwen 2.5 14B model, according to research from HonestAGI.
The accusations include concerns about discrepancies in how Huawei described its training process and adherence to open-source licensing rules, raising potential copyright issues.
Read at Developer Tech News
[
|
]