⏳ In Brief
-
Huawei’s Noah’s Ark Lab rejects claims its Pangu Pro MoE model copied Alibaba’s Qwen 2.5 14B
-
Allegations stem from HonestAGI’s GitHub post citing “extraordinary correlation”
-
Huawei insists Pangu Pro MoE was independently built and trained on Ascend chips
-
The lab confirms the use of open-source code under proper licensing
-
Alibaba has not publicly responded; the accuser remains anonymous
🧠 Lab Officials Reject Copying Allegations
Huawei’s Noah’s Ark Lab has come out swinging against accusations that its flagship Pangu Pro MoE language model was derived from Alibaba’s Qwen 2.5 14B model.
The controversy erupted when an anonymous GitHub user, HonestAGI, published a technical critique claiming an “extraordinary correlation” in outputs and architecture between Pangu Pro MoE and Qwen 2.5 14B.
“We categorically deny any suggestion that Pangu Pro MoE was built on or trained from Alibaba’s Qwen,” said a lab spokesperson.
Huawei emphasizes that Pangu Pro MoE’s design and training pipeline were developed entirely in-house.
🔍 Open-Source Components and Licensing
Huawei acknowledges that open-source libraries were used in Pangu Pro MoE’s preprocessing and tooling pipelines.
The lab confirms it complied fully with MIT, Apache, and other open-source licenses, attributing and packaging code according to each project’s terms.
“All third-party code was incorporated under clear license terms; no proprietary model weights were reused,” the spokesperson added.
Proper licensing and attributions are central to Huawei’s AI development governance.
🌐 Industry Context and Competitive Stakes
This clash emerges amid an intensifying rivalry among China’s AI giants. Huawei, Alibaba, Baidu, and others are each vying to lead in large-scale language models and agentic AI services.
Huawei’s Pangu Pro MoE, open-sourced on GitCode in June 2025, targets enterprise and government use cases, offering a mixture of experts architecture for high-throughput tasks.
By contrast, Alibaba’s Qwen series focuses more on consumer and developer ecosystems, integrating into e-commerce and cloud services.
Allegations of copying threaten reputations and could slow industry collaboration on open innovation.
🛡️ Next Steps and Industry Implications
Huawei plans to commission an independent audit of its codebase and model training logs to prove the originality of its code definitively.
Meanwhile, Alibaba has stayed silent, and HonestAGI’s GitHub notice was swiftly removed, leaving technical details scarce and fueling speculation.
Industry observers warn that such disputes may spur calls for standardized AI provenance tools to track datasets and the lineage of architecture.
“In a maturing AI field, transparency around model provenance is non-negotiable,” stated Dr. Mei Lin, an AI ethics researcher.
📌 Conclusion
Huawei’s forceful denial of model copying underscores the high stakes in the AI model arena.
As Pangu Pro MoE and Qwen continue to evolve, both labs must strike a balance between open collaboration and stringent intellectual property safeguards—or risk eroding trust in China’s burgeoning AI ecosystem.
📈 Trending News
7th July 2025:
- Samsung’s AI Chip Delays Cause Q2 Profit to Plunge 39%
- Grok’s ‘Upgrade’ Targets Jewish Executives in Hollywood and the Democratic Party
- How AI VTubers Like Bloo Are Making Millions on YouTube Without Ever Showing Their Face
- Unity’s AI Demo Accidentally Generates Mickey Mouse on Live Stream
- Swedish No-Code AI Startup Lovable Nears $2B Valuation
For more news and insights, visit AI News on our website.
