Huawei's version of ChatGPT will be released in early July, called "Pangu Chat"

Source: Titanium Media

Author: Lin Zhijia

**News on June 3, Titanium Media App recently learned exclusively from Huawei that Huawei will release a multi-modal 100-billion-level large-scale model product called "Pangu Chat" that directly matches ChatGPT. **

**It is expected that Huawei Pangu Chat will be released and internally tested at the Huawei Cloud Developer Conference (HDC.Cloud 2023) held on July 7 this year. The product is mainly for To B/G government and enterprise customers. **

This means that in the domestic large-scale model arms race, after Ali and Baidu, another important technology giant has entered the game. Based on Huawei's technical capabilities, Pangu Chat is expected to become the ChatGPT product with the strongest technical capabilities in China, and Huawei's ecological industry chain enterprises will also benefit from it.

**It is reported that **Pangu Large Model was successfully established in Huawei Cloud in November 2020. For the positioning of the Pangu large model, Huawei's internal team has established three key core design principles: first, the model must be large enough to absorb massive amounts of data; The generalization ability can truly be applied to work scenarios in all walks of life.

In April 2021, the large model of Pangu was officially released to the public. Among them, the Pangu NLP large model is the first Chinese pre-training large model with 100 billion parameters, and the CV large model has reached 3 billion parameters for the first time. In April 2022, Huawei Pangu upgraded to 2.0, released a hierarchical development plan (LO, L1, L2), and built an industrial-grade Pangu model.

According to the presentation PPT information of Huawei Cloud executives, at present, the basic layer of Huawei's "Pangu series AI large model" mainly includes NLP large model, CV large model, and scientific computing large model, etc., and the upper layer is Huawei's industry large model developed with partners. .

Tian Qi, chief scientist in the field of artificial intelligence at Huawei Cloud, said in April this year that in 2022, Huawei has just delivered a 100-billion-parameter large model in Arabic, covering areas such as intelligent document retrieval, intelligent ERP, and small-language large-scale models covered by the Pangu NLP large-scale model. . In terms of application, the Pangu model can be applied in many fields such as intelligent customer service, machine translation, speech recognition, etc., providing AI technical support.

Last year, Huawei applied the Pangu large model in the vertical industry field. An example is that based on the Pangu AI large model, Huawei built the first "artificial intelligence training center based on a large model" in the field of mines to realize the "industrialized" development of Al applications, which not only reduces costs - 80% more than manual verification, The cost per ton is reduced by 8 yuan (annual production capacity is 2.6 million tons), and the benefit is increased - the anti-shock pressure is 80%+ lower than that of manual verification, and the efficiency is improved - the efficiency of smart customers is increased by 30%.

Compared with other manufacturers, Huawei's advantage may lie in its complete industrial chain and strong computing power deployment capabilities. According to the previous disclosure of Zheshang Securities, when ** was training the Pangu large model with 100 billion parameters, the Huawei team used more than 2,000 Shengteng 910 chips for more than 2 months of data training capabilities. According to Huawei internally, more than 4,000 GPU/TPU cards are used for large-scale model training every year, and the computing power cost of large-scale models in three years is as high as 960 million yuan. **

According to the research report released by Soochow Securities on March 27, the advantage of Huawei's Pangu large model lies in its talent reserve and independent controllable computing power, and it is expected to become the leading large model in China. Its ecological industry chain targets are expected to usher in accelerated development, including Tuowei Information, Sichuan Changhong, Kylin Software (China Software), Tongxin Software (ArcherMind Technology), Kylin Principal and other Huawei ecological companies. Guosheng Securities believes that Huawei Pangu is the first multi-modal 100 billion-level large-scale model, which is expected to empower all industries.

According to the data released by Huawei, the parameters of Huawei's Pangu-Σ large model are at most 1.085 trillion, which is developed based on Huawei's self-developed MindSpore framework. **On the whole, the PanGu-Σ large model may be close to the level of GPT-3.5 in terms of dialogue. **

It is worth noting that because the Pangu model has very large parameters, the training cost is relatively high. Although the Pangu model has high accuracy and quality in Chinese language processing, it still has certain limitations in semantic understanding, especially in dealing with complex language structures and semantic reasoning. At the same time, the large model of Pangu Chat will also face fierce competition in the industry, and it needs to continuously optimize and improve the performance and quality of the model in order to maintain its leading position in the market. (This article is the first release of Titanium Media App, author｜Lin Zhijia)