Its name is ＂Zhaoyan＂! Chongqing’s first self-developed large language model was officially released.

▲ On-site signing. Reporter Zhang Yizhu photo

On July 13th, Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University officially released the "Zhaoyan" large model in the West (Chongqing) Science City, which is also the first large language model independently developed in Chongqing.

With the global fire of ChatGPT, the big model has become the hottest topic at present, and 2023 will also become the year of big model generate. Statistics show that hundreds of large models have been released around the world, and more than 80 large models have been released in China since this year.

"The previously released general model represented by ChatGPT has some pain points in the industry, such as lack of domain knowledge, inaccurate system calls, and high deployment costs." Jin Shi, dean and chief scientist of Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University, said that for example, the general big model is like a "generalist", but due to the lack of professional knowledge and industry data accumulation, its industry depth is not enough, and the system call is not accurate, so when answering some highly professional questions, it may be "talking nonsense in a serious way".

"Zhaoyan" is committed to building a big model of the industry, just like a "professional". It is developed by Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University based on the world’s leading AI big model technology. By accumulating massive data, supplemented by unique data processing and AI training technology, it can achieve accurate and professional text output, and can realize the ability to select and call applications in natural language, helping to create personalized intelligent products.

"Does Jiangzhou in this Jiujiang official. My blue sleeve was wet mean Chongqing?" On the same day, the press conference demonstrated "Zhaoyan" and compared it with well-known large models in the industry.

The Chongqing Daily reporter saw that the "Zhaoyan" answer not only introduced the historical relationship between the name Jiangzhou and Chongqing, but also did not confuse Chongqing and Jiujiang, knowing that Jiangzhou is Jiujiang now. But another big model gives the wrong answer.

▲ Jin Shi, Dean and Chief Scientist of Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University, introduced the "Zhaoyan" big model. Reporter Zhang Yizhu photo

"’mega-words’ means’ mega-words accumulate and speak out in chapters’." Jin Shi said that this not only expresses that the process of creating a large model involves the collection and processing of massive data, that is, "mega-text accumulation", but also expresses that the large model can achieve high-precision and professional-level text output, that is, "speak out" and let people get the information they really need.

In addition, the research and development of "Zhaoyan" also focuses on miniaturization deployment. Through distillation learning, model compression and special reasoning chip, the cost of model training and reasoning is greatly reduced. For specific tasks in the industry, the quality of answers is not reduced, while the cost of model call is only one-fourth of that before miniaturization, so that more small and medium-sized enterprises can also apply large models and make model-based products more competitive in the market.

"To sum up,’ Zhaoyan’ has three characteristics: high precision, high efficiency and low cost, and better empowers manufacturing, medical care, finance, cultural tourism, law, service and other industries." Introduction of Jinshi. For example, in the medical field, "Zhaoyan" can provide medical assistance and improve the quality and efficiency of medical services by deeply learning medical data.

"In the past, medical services were limited by doctors’ professional knowledge and experience, and the consultation response speed was slow and the service quality was limited." For example, he said that "Zhaoyan" can deeply learn medical records and symptom descriptions, provide comprehensive and accurate diagnosis and treatment suggestions, and improve the response speed and quality of medical services.

▲ The press conference site. Reporter Zhang Yizhu photo

At the press conference, Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University also signed contracts with Huayuan Computing Technology (Shanghai) Co., Ltd., Chongqing CCID Engineering Consulting Co., Ltd., Chongqing Yucun Technology Co., Ltd., China Mobile Communications Group Chongqing Co., Ltd., Beijing last romance Media Technology Co., Ltd. and Chongqing Saibao Industrial Technology Research Institute Co., Ltd., which will give full play to their respective advantages and promote the application of large-scale model technology in different industry scenarios such as manufacturing, finance, cultural tourism, education and medical care.

In the future, Chongqing Institute of Artificial Intelligence of Shanghai Jiaotong University will continue to optimize and upgrade in the vertical field of the industry according to the demand of Chongqing’s industrial layout, and apply "Zhaoyan" large model products to various fields to help the construction of the western (Chongqing) science city and the development of Chongqing’s digital economy.

Chongqing Daily client reporter Zhang Yizhu

Original title: Its name is "Zhaoyan"! Chongqing’s first self-developed large language model was officially released.

Editor: Rao Zhimei

Editor: Wu Zhonglan

Audit: Feng Fei