Shulou(Shulou.com)11/24 Report--

2023 Tencent Global Digital Ecology Conference "Industry Big models and Intelligent applications" was held on September 7. At the meeting, Tencent Cloud and Xintong jointly released the "Research report on the Standard system and capability Architecture of Industry large models". With reference to Tencent Cloud's practice and experience in the industrial application field of industry large models, it provides a standardized process and comprehensive evaluation method for customers to build industry large models. In order to better help industry partners to build their own large models.

At the ecological conference, Tencent also announced the progress of Tencent mixed meta model for the first time, and announced the upgrade of MaaS capability from the three dimensions of technology base, platform capability and intelligent application. Based on the self-developed Tencent mixed meta model, it also supports more than 20 mainstream open source models in the industry, helps customers build exclusive large models and intelligent applications, accelerates the landing of the model industry, and promotes the "performance revolution".

With the gradual increase of the scale of artificial intelligence model and the significant improvement of performance, various industries have ushered in new opportunities for development. Compared with the general model, which is more focused and professional, the industry model has a wide range of applications and business innovation value in specific industry scenarios. However, when the large model of the industry falls into the real scene, customers are still faced with some problems, such as high input cost, few computing resources, poor data quality, scarcity of professionals and so on. Based on the pain points of industry users, Tencent Cloud and Communications Institute have carried out research on the standard system and capability architecture of industry large models to help enterprises solve the application problems of industry large models and improve the efficiency of using large models for digital transformation.

Establish standardization process to guide enterprises to build large model ability

The construction of industry large model puts forward higher requirements for industry data, computing resources and deployment operation and maintenance, which has a certain threshold for many enterprises, and there is a lack of guidance scheme for industry large model construction method and process at present. Enterprises do not know how to build industry large models. Based on this problem, Tencent Cloud and CICT set up standardization processes for vertical industry models to guide enterprises to build large model capabilities.

To build a large-scale model of the industry, we first need to evaluate the business needs and resources, including business needs assessment, computing layer assessment, algorithm layer evaluation, data layer evaluation and engineering layer evaluation.

"Business needs assessment" needs to be combined with actual scenarios and requirements, define business data, business indicators, and business service methods, and determine the selected basic model and fine tuning algorithm through the evaluation of computing layer, algorithm layer, data layer and engineering layer. According to data security, privacy, data quality and other issues, determine the platform adaptation and training flow.

The co-construction of industry data and large model includes clear scene objectives, model selection, training environment construction, data processing and so on. For the co-construction of industry data and large models, first of all, it is necessary to make clear the practical application scenarios and model evaluation objectives of industry large models, select the appropriate basic large models in the model base, and through the construction of hardware environment and software platform, the data is processed into the data needed for model training, and finally, the training and co-construction of the industry large model in a specific scene is carried out. It includes the key links such as data upload training platform, preparation of training code and environment, initiation of model training task, storage of model after training, evaluation of model after training and so on.

Fine tuning and optimal deployment of large models in the industry, including model fine tuning, model evaluation, model retraining and optimization, model joint tuning and deployment, model application operation and so on. According to the evaluation of business needs and resources and the co-construction of industry data and large models, the training of large models is fine-tuned by using industry-specific scene data prepared in advance, and the generated models are evaluated and optimized for multiple rounds of iterative training. Then it carries on the reasoning service deployment, service configuration, application service joint adjustment to the model, and provides the monitoring, management, operation and maintenance capabilities to the online model service.

Perfect the standard system and establish the ability evaluation index in an all-round way

With the rapid rise of large models, the industry has also realized that the development of innovative technology is inseparable from the promotion of standardization, and the construction and improvement of the standard system of large models is of great significance to promote the application of large models and accelerate the integration of large models and industry applications. it is of great significance to promote the industry to develop in the direction of standardization, compliance and scale.

As an important research institution in the field of national information and communication, ICT has been paying close attention to the development and application of AI technology, especially the innovation and promotion of large model technology. Focusing on the key links such as "construction, use and management" of the basic large model in the actual demand, CICT took the lead in building and improving the standard system of the large model from many directions, and Tencent participated in the whole standard work in the whole process and deeply, focusing on the engineering application. Around the five aspects of model development, model capability, model operation, model application and model credibility, a standard index system is formed to comprehensively evaluate the comprehensive level of the large model.

The standards and evaluation indicators of large models in finance, culture, tourism, government affairs, media, telecommunications, education, automobile and other industries play a key role in promoting the high-quality development of industry models, enhancing user experience and improving the level of digital intelligence. On the one hand, industry large model standards can be used to guide third-party evaluation institutions to evaluate and accept the capabilities of large models in specific industries, and on the other hand, evaluation indicators can examine the coverage of the model in various application scenarios. and performance in data processing, data compliance, data privacy and security.

"We hope that the research on the standard system and capability architecture of industry large models can help industry partners solve the problems of how to" apply "large model capabilities" to their own industries and scenarios, and can more effectively solve the cost, data, security, compliance and other large model landing problems. through the establishment of a standard system to provide guidance and support for the construction and application of large models in the industry. " Jiang Bo, vice president of legal affairs of Tencent, mentioned.

While more than 20 industries have landed large models, Tencent Cloud has been actively participating in and promoting the construction of industry large model standards. As early as 2020, Tencent was elected as a member and deputy secretary-general of the artificial Intelligence Sub-Committee of the National Beacon Commission. Tencent is a core member in the process of promoting new infrastructure, including artificial intelligence. Tencent has undertaken a lot of standard-setting work and technical leadership.

Not long ago, Tencent Cloud and China Information Communications Institute jointly launched a joint promotion plan for industry large model standards, and jointly led the development of China's first financial industry large model standards, providing important support for the intelligent and high-quality standardization development of the financial industry.

He Baohong, director of the Cloud Computing and big data Research Institute of the China Academy of Information and Communications, said that with the rapid development of artificial intelligence technology, industry large models have become an important driving force for industrial upgrading. In order to standardize and lead the healthy and sustainable development of the industry large model, it is necessary to construct the standard system of the industry large model in order to scientifically evaluate the technical capability and application effect of the large model. I hope the results of this research will contribute to the development of large model industry and ecological construction.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.