Skip to content

Huawei insists on the homegrown nature of its AI models, following concerns voiced by informants.

Open-source GitHub account claims remarkable associations between Huawei's product and Qwen-2.5 14B of Alibaba.

Huawei vigorously asserts domestic development of AI models, countering concerns raised by insiders
Huawei vigorously asserts domestic development of AI models, countering concerns raised by insiders

Huawei insists on the homegrown nature of its AI models, following concerns voiced by informants.

In the competitive world of artificial intelligence (AI), allegations of plagiarism have surfaced regarding Huawei's latest open-source language model, Pangu Pro MoE. The claims, made by an account on GitHub named HonestAGI, suggest a striking similarity between Huawei's model and Alibaba's Tongyi Qianwen (Qwen-2.5 14B).

However, Huawei has firmly denied these allegations. In a statement issued by Huawei's Noah's Ark Lab, the company emphasised that the Pangu Pro MoE was developed and trained independently from scratch on Huawei's own Ascend hardware platform. The statement also rejected the assertion that their model was incrementally trained or copied from other providers.

Despite using industry-standard open-source practices and properly licensed open-source components, Huawei maintains that the Pangu Pro MoE embodies original architectural and technical innovations. The company reaffirmed their commitment to competing globally with their AI technology.

The controversy arises amidst intense competition in China’s AI market and efforts to reduce dependence on Nvidia GPUs due to export restrictions. Huawei promotes Ascend chips as alternatives.

Noah's Ark Lab, the unit responsible for Pangu model development, clarified that the Pangu Pro MoE open-source model was developed and trained on Huawei's Ascend hardware platform. The original repository uploaded by HonestAGI has since been removed, but a brief explanation remains.

Huawei is considered a key symbol of China's resilience against US tech sanctions. The company operates a lab specializing in large language models (LLMs), and the Pangu Pro MoE model is the latest open-source model developed by Huawei's Noah's Ark Lab.

It's worth noting that Alibaba Group Holding, the owner of the South China Morning Post, also develops open-source LLMs, posing competition to Huawei. Open-source LLMs developed by other entities, such as DeepSeek, are also gaining traction in the market.

In summary, there is no public concrete proof of plagiarism, only allegations based on technical similarity analyses, which Huawei has officially and strongly rejected. The company stands by its claim that the Pangu Pro MoE was developed and trained independently.

Technology plays a significant role in the rebuttal by Huawei, as they emphasize that their Pangu Pro MoE was developed and trained independently on their own Ascend hardware platform, distinct from any other providers, including Alibaba. The artificial-intelligence-driven model, a testament to Huawei's original architectural and technical innovations, remains a point of pride for the company in their global competition.

Read also:

    Latest