The AI agent creation and distribution platform "Tencent Metaware" was launched, opening the application experience.
The landing of large models has accelerated, and "industrial practicality" has become a development consensus.
On May 17, 2024, Tencent Cloud Generative AI Industry Application Summit was held in Beijing, and a series of progress in large-scale model research and development and application products was announced.
Tencent’s mixed-element model capability has been continuously upgraded, and several versions of models, hunyuan-pro, hunyuan-standard and hunyuan-lite, have been opened to the outside world through Tencent Cloud, meeting the model needs of enterprise customers and developers in different scenarios and landing the best cost-effective model scheme.
Tencent Cloud’s big model knowledge engine, image creation engine and video creation engine are released, creating a native tool chain in the big model era, simplifying data access, model fine-tuning and application development processes through PaaS services, helping enterprises to develop AI native applications with big models more efficiently and simply, and quickly access production scenarios.
Taking "industrial practicality" as the core strategy of developing a large model, Tang Daosheng, senior executive vice president of Tencent Group and CEO of Cloud and Smart Industry Group, said that by building a high-performance model, an efficient tool platform, a highly agile scene application, a highly available computing infrastructure, and a strong and secure model environment, the AI closest to the industry will be built.

Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud and Smart Industry Group
Tencent Hybrid Open 256k Long Text Model on Cloud, Opening Agent Ecology
Powerful general-purpose large models and low-threshold development tools can improve the productivity of developers and promote the ecological development of large models.
At this industry summit, Tencent Hunyuan introduced its multi-dimensional LLM model matrix, in which the largest model has been expanded to the scale of trillions of parameters, and has layouts under different parameters such as 1B, 3B, 7B and 13B.
After the upgrade, Tencent Mixed Element takes the lead in adopting the mixed expert model (MoE) structure in China. Compared with the previous generation, the overall performance of the model has been improved by 50%, and some Chinese abilities have been tied with GPT-4. In the performance of answering "up-to-date" questions, the abilities of mathematics and reasoning have been greatly improved.
On Tencent Cloud, the mixed-element model also provides model services in various sizes, such as trillion parameter hunyuan-pro, billion parameter hunyuan-standard, and billion parameter hunyuan-lite. At present, it has been fully open to enterprises and individual developers.

Tencent’s mixed model service is open to the outside world through Tencent Cloud.
Among them, hunyuan-standard has recently launched a long text model supporting a 256k long context window, which has the ability to process a long text with more than 380,000 characters at a time, and has shown strong performance in reading and understanding long documents and analyzing large-scale data. It can provide strong work support for professionals in finance, medical care, education, travel and other industries, and significantly improve work efficiency.
In terms of multi-modal capability, the mixed-element large model also continues to be iteratively upgraded. In the field of raw graphics, the infrastructure of Tencent’s mixed-element raw graphics has been fully upgraded to the DiT architecture of Sora, which supports bilingual input and understanding in both Chinese and English, and has the ability of multi-round drawing, and the evaluation results are leading in China; In the field of raw video, Tencent Hybrid supports various video generation capabilities such as Wensheng video, Graphic video, Graphic video and Video video, and has already supported the generation of 16s video. On the 3D level, Tencent Hybrid has laid out a text/image 3D, and a single image can generate a 3D model in 30 seconds.
According to Sullivan’s evaluation results, Tencent’s general basic ability and professional application ability are in the leading echelon of domestic large models, higher than the average of international large models. The report of the authoritative evaluation agency SuperCLUE also shows that Tencent’s mixed-element big model ranks in the first echelon of domestic big models, and is in a leading position in basic and scene applications, and is located in the quadrant of outstanding leaders.
At the meeting, Jiang Jie, vice president of Tencent Group, announced that Tencent’s mixed model will embrace open source. Previously, the mixed-meta-graph model has been fully open source, and it has attracted the attention of more than 1,000 developers on Github in just 3 days. Tencent’s mixed-element MoE models of various sizes will soon be open to the outside world, which can support diverse deployment scenarios such as mobile phone, PC and cloud/data center respectively.

Jiang Jie, Vice President of Tencent Group
As a practical big model, the mixed-element big model has been tested in more than 600 internal businesses and scenarios of Tencent, and it continues to iterate in Tencent’s rich ecology. Based on the mixed-meta model, WeChat Reading has launched new functions such as AI book inquiry and AI outline, which greatly improves the reading efficiency and experience of users. Tencent’s customer service team upgraded the intelligent customer service system based on the mixed model, and created a fine-tuning model of the vertical field of intelligent customer service, which greatly improved the accuracy of understanding the intention of intelligent dialogue and the fluency of multiple rounds of questions and answers. Compared with the traditional small model, the accuracy of the following text was improved by 38%. Its artificial customer service assistant was applied in multiple game customer service scenarios, and the daily average user request reached 1.5 million times. Tencent conference is based on the AI assistant launched by Hunyuan, which can immediately answer questions inside and outside the conference and greatly improve the efficiency of the conference. In the past four months, the daily call volume of Tencent Conference AI Assistant has increased by 20 times. Collaborative SaaS products such as enterprise WeChat and Tencent Document are also fully connected to Tencent Hybrid. Tencent Advertising has launched a one-stop AI advertising creative platform based on Tencent Hunyuan, which helps to improve the efficiency of advertising production and delivery.
Jiang Jie said that externally, Tencent Hybrid will also open the agent ecology and launch the one-stop AI agent creation and distribution platform "Tencent Meta Device". In the future, users can not only create exclusive AI agents on the platform, use Tencent’s official plug-ins and knowledge base, but also distribute these agents to QQ, WeChat customer service, Tencent Cloud and other channels with one click.

Tencent yuan organ network is open for trial application
Tencent Cloud released three AI big model engines to build a knowledge service application in 5 minutes.
With the big model technology as the core, artificial intelligence has become the key driving force for the digital development of enterprises. Research shows that more than 60% of China enterprises plan to deploy generative AI in the next 12 to 24 months.
But how to find the right scene, deploy it quickly, and shorten the distance from basic model to industrial application?
In the past year, in the process of serving industrial customers, Tencent found that the demand for models in the industry is constantly changing. On the one hand, with the diversification of industrial information carriers, the demand of the model is not only to deal with simple words, but also to deal with pictures, videos and other information. The ability competition of large-scale model has expanded from a single literary creation to multi-modal ability such as literary creation diagram, literary creation video, graphic creation diagram and graphic creation video. On the other hand, in the environment of reducing costs and increasing efficiency, enterprises have higher requirements for "cost performance", and expect to use simpler large-scale model tools to accelerate application development, realize rapid production and meet the sustainable input-output ratio.
In order to better meet these needs, Tencent Cloud has launched a new native tool chain for large models, with three PaaS products —— large model knowledge engine, large model image creation engine and large model video creation engine to help enterprises improve quality and efficiency in knowledge services, image and video creation.

Tencent Cloud released three AI engine tools to lower the threshold of model application.
Among them, the big model knowledge engine focuses on the enterprise knowledge service scenario, and integrates OCR document analysis, vector retrieval, large language model, multi-modal big model and other technologies based on the RAG (Search Enhanced Generation) technical architecture, creating a "low threshold" and "high efficiency" model application development platform for enterprises. Through the "modular" application template, enterprises can develop a knowledge service application in natural language in 5 minutes.
At present, the knowledge engine of Tencent Cloud Big Model has been applied in many industries such as government affairs, finance, education, travel and retail. In the financial industry, Center Huibao has developed an efficient Huimin think tank for insurance agents, and with the assistance of large-scale model technology, it automatically generates product knowledge questions and answers and soothing words, achieving a per capita efficiency increase of 50%. In the education industry, Henan Digital Education Development Co., Ltd. used the knowledge engine to import millions of primary and secondary school textbooks in Henan Province, and arranged the knowledge to create a 7×24-hour all-weather model knowledge assistant.
Within Tencent, a number of SaaS applications are upgraded by relying on the knowledge engine. In the customer service scenario, enterprises point to a large model text robot for customer service, and access the large model multi-round task engine to perform tasks such as bill inquiry and return, and the configuration cost is reduced by 50% compared with the traditional text robot. In the service scenario of digital people, after accessing the knowledge engine of the big model, digital homo sapiens can better understand and identify users’ intentions, and use the big model to generate more professional and personalized answers. In the enterprise knowledge service scenario, Tencent enjoys the combination of knowledge engine, providing "intelligent writing and generation" capability on the knowledge production side and "intelligent question and answer" capability on the knowledge consumption side, so as to make enterprise employees’ knowledge production and learning more efficient and enhance organizational ability.
In addition to the big model knowledge engine, the image and video creation engine will comprehensively improve the efficiency of material generation through the big model. The "Image Creation Engine" is based on Tencent’s self-developed image creation model, which outputs high-quality AI image generation and editing capabilities, and provides corporate customers with AI photo, line drawing, image stylization and other capabilities. For example, in the design scene, enterprise customers use the function of "line drawing generation" to upload product line draft design drawings, and then quickly generate physical design drawings through prompt words and parameter settings, greatly shortening the creation and production cycle.
Based on multi-modal algorithm technology, Video Creation Engine outputs high-quality video generation and processing capabilities, and provides video translation, video stylization, canvas expansion and other functions. Facing the demand of enterprises going to sea, "video translation" helps enterprise customers to translate the original video into multi-language video output with one click, quickly put it into overseas markets and seize the sales opportunity.

Wu Yunsheng, Vice President of Tencent Cloud, Head of Tencent Cloud Intelligence, Head of Youtu Lab and Head of Tencent Enterprise.
Wu Yunsheng, vice president of Tencent Cloud and head of Tencent Cloud Intelligence, said that Tencent Cloud started from the actual needs of the industry, built a native tool chain in the era of big models, and relied on three AI big model engine tools to realize data engineering, model fine-tuning and application development. Streamlined processes help enterprises use big models more efficiently and conveniently.
Computing power and safety double base upgrade, escort production AI development
Generative AI drives "intelligent emergence", which brings growth opportunities for enterprises and new security challenges. In the process of industrial practice, Tencent found that there are two major obstacles for enterprises to embrace generative AI, namely, the shortage of computing resources and security concerns.
Safety compliance is the bottom line for enterprises to apply artificial intelligence technology. Based on the accumulation of security technology for more than 20 years, Tencent Security upgraded and launched a systematic security solution for AIGC scenarios.
In terms of data security, Tencent Security has launched a full-link data security solution. Through tools such as key management system, fortress machine and data security governance center (API security monitoring), it escorts the data security of enterprise model training, fine adjustment, release and operation throughout the life cycle, helping enterprises to protect sensitive data and ensure the safety and compliance of data collection.
In terms of content security, the content generated by the big model often encounters some unknown risks, such as false information, content infringement, induced risk and personal privacy. Tencent Cloud Tianyu AIGC content compliance solution, through the five service systems of expert service, data service, copyright service, computer audit service and CEM (customer experience management) service, solves the content security challenges such as model training, content generation and post-event operation of AIGC applications in the whole process.
At present, Tencent Cloud Tianyu has escorted a number of AIGC formats, covering scenes such as AI Q&A, digital people, creative assistants, literary drawings, code generation, entertainment and social interaction, and AI customer service.
At the computing power level, Tencent Cloud provides a one-stop AI infrastructure for the industry training model. Tencent Cloud has built a computing cluster that can support more than 100,000 cards for parallel computing and is compatible with various GPU ecosystems through its self-developed 3.2T communication bandwidth and unified access layer capability. Tencent Cloud also launched the first AI native vector database in China, which can support up to 100 billion-level vector scale. It is the first product in China to pass the capability evaluation of ICT vector database.
Generative AI ecological plan is released to build a prosperous ecological drive for industrial intellectual change.
The landing of large-scale model industry is a vast market, but also a complex process, which requires the whole industrial chain of large-scale model manufacturers, physical industries and ecological partners to attack. Since 2023, Tencent Cloud has worked closely with 1,500 partners, relying on leading and rich generative AI products, serving more than 20,000 corporate customers, and initially building an ecosystem around generative AI products.
At the summit, Tencent Cloud officially launched a generative AI ecological plan. Yang Chen, vice president of Tencent Cloud and head of industrial ecological cooperation, said that Tencent Cloud will focus on strengthening the generative AI technology and platform base, unite thousands of solution providers in the future by opening platform capabilities and services and providing technical and market support, cultivate thousands of service providers and tens of thousands of agents, and jointly promote the generative AI technology to penetrate into the whole industry chain and accelerate the intelligent upgrading of the industry.

Tencent Cloud and 17 partners released a generative AI ecological plan.
In terms of open capabilities, Tencent Cloud will open up full-generated AI products for solution providers, as well as platforms and atomic capabilities such as PaaS, aPaaS and iPaaS to help partners form differentiated and competitive solutions on the application side. In terms of service opening, Tencent Cloud will focus on cultivating thousands of professional service partners of generative AI products, and through systematic capability certification, partners will form a capability matrix of full-process services to improve quality delivery for the industry. In terms of technology and market support, Tencent Cloud provides partners with special testing funds for seven major generative AI core products, such as knowledge engine, vector database and digital homo sapiens, and arranges 100 large model architects, product experts, algorithm experts and data experts to accompany them, helping partners to accelerate the landing and replication of customer scenarios. At the same time, the establishment of generative AI marketing plus multiplication funds will effectively help partners explore the market.
At the meeting, Tencent Cloud and Gartner also released the Research Report on the Landing Path of Generative AI Industry (hereinafter referred to as "Report"), providing enterprises with a scenario matrix and a roadmap for the landing of generative AI applications to help enterprises solve challenges such as scenario value and landing feasibility.
Tang Daosheng said that facing the smart future, Tencent will always adhere to the strategic direction of "industrial practicality", persist in using technology to solve practical problems, and will also adhere to ecological opening up and work with industries and partners to help the industry meet the smart future. "







