On April 24, during the 2024 Beijing Auto Show, Tencent Cloud released the "Global Intelligence" solution for large-scale models in the automotive industry, bringing users a full chain of services from models, computing power, AI engineering platforms to AI applications, aiming to comprehensively cover all aspects of R&D, production, marketing, and help car companies improve efficiency in an all-round way.
In terms of model capabilities, Tencent Cloud's automotive industry large model adds massive professional data of the automotive industry on the basis of the full-link self-developed hybrid large model, performs pre-training, fine-tuning vertical tasks in the automotive field, and reinforcement learning, and performs well in Chinese reading comprehension, end-to-end Q&A, and automotive industry-related tasks.
In addition, the new upgrade of the Tencent Cloud TI platform tool chain, combined with the continuous expansion and update of the mainstream open source model libraries supported by the platform, has further improved the efficient training and application capabilities of large models, accelerating the intelligent development of the automotive industry.
Focus on scenarios such as digital human innovation marketing, and help car companies apply full-link models
Tencent Cloud has been committed to promoting the application of AI large models in automotive and other industries, and has continuously improved model performance by strengthening the base model, optimizing the tool chain, enhancing computing power support, and in-depth business scenarios, helping users build exclusive automotive large models and intelligent applications, so as to accelerate the practical application and implementation of large model technology in the automotive industry.
For example, in the process of automobile marketing and sales, large models are being integrated into the whole journey marketing plan of users. Relying on the AI model, it can not only generate a realistic virtual digital human image of 1-minute video material in 1 hour, but also be broadcast live by the digital human, interact with users in real time, and exert real productivity in marketing and service scenarios. Compared with traditional heavy investment advertising, digital human live broadcast has the advantages of fast production, low cost and high refresh rate, which can quickly adapt to the new media marketing environment and efficiently generate high-quality and high-quantity promotional materials.
In the after-sales service scenario, Tencent Cloud's large model knowledge engine integrates car manuals and other materials to empower digital humans with the ability to generate personalized answers in real time, effectively solving customers' questions in vehicle use and maintenance, and significantly improving the customer service experience. The knowledge engine adopts the LLM+RAG framework, and integrates OCR, MLLM, vector database and content security technologies, especially in the processing of automotive manuals and knowledge questions and answers with mixed graphics, text and tables, reaching the industry-leading end-to-end application level.
The realization of these application effects benefits from the in-depth optimization and technological innovation of Tencent Cloud's large model in the automotive industry. Based on the base of the self-developed hybrid model, Tencent Cloud's industry model is particularly good at logical reasoning and long text creation. At the same time, under the fine-tuning of massive professional data in the automotive industry, it has also been specially optimized for problems such as hallucination, rejection, poor response to multiple questions, and insufficient information extraction ability.
The upgrade of development tools has also further promoted the in-depth application of AI large models in the automotive industry. Based on the needs of the industry and users, the Tencent Cloud TI platform has undergone a comprehensive tool chain upgrade. The platform presets 140 datasets of different types of large model fine-tuning task scenarios, covering data cleaning, optimization, filtering, and enhancement, forming a complete data processing process. This upgrade greatly improves the efficiency and quality of data processing, and can complete the fine-tuning of large models with one click, greatly reducing the threshold for users in the automotive industry to access and use AI large models.
In addition, Tencent Cloud has adopted a new generation of HCC high-performance computing clusters to achieve ultra-large computing power of 100,000 GPUs, which can quickly train trillion-parameter large models. Combined with the 3.2 Tbit/s communication bandwidth of the self-developed Xingmai network, the GPU utilization rate is increased by 40%. At the same time, the self-developed training and inference framework saves 50% of the computing cost in the training of 100 billion parameter large models.
Implemented in multiple industries such as finance, medical care, and government affairs, and continued to improve intelligent services and operational efficiency
The rapid development of AI large models has brought a new increment to the cloud market and accelerated the process of cloud intelligence integration. Since the official release of the industry model solution in June last year, Tencent Cloud has continuously iterated model capabilities through the Tencent Cloud TI platform to provide customers with one-stop model services based on the self-developed Tencent Hybrid Model and supporting mainstream open source model applications in the industry.
At present, Tencent Cloud's TI platform has launched mainstream models in the industry, including Llama3, Llama2, Falcon, and Dolly, which support direct deployment and calling, simple application processes, and low-code operations throughout the process. Enterprises and developers can flexibly select various large models according to the business needs of different subdivision scenarios to reduce the cost of using models.
In addition to the automotive industry model, Tencent Cloud, together with ecosystem partners, has combined the model technology with more than 20 industries to provide more than 50 industry model solutions to help customers improve quality and efficiency. In addition, it actively participates in the construction of industry large-scale model standards to help the high-quality development of the real industry.
As one of the first vertical fields where large models are implemented, the financial industry has abundant data resources and a wide range of practical scenarios. China UnionPay has built a one-stop AI development platform through the Tencent Cloud TI platform, effectively lowering the threshold for AI capacity building in financial services. In the document recognition and entry scenario, the platform supports UnionPay to quickly realize the customized recognition of certificates, bills, and texts by integrating the OCR large model, so as to automate business processes, improve efficiency and save costs. At the same time, Jinteng Technology adopts the mode of combining public and private domains to realize hot spot analysis, market summary, stock market closing and other applications in the public cloud environment, and builds an enterprise-specific financial model in the private cloud environment to help institutions achieve intelligent upgrades in business scenarios such as product information inquiry, business guidance, investment advice, and post-investment management.
In terms of medical and health care, Shanghai Renji Hospital took the lead in using Tencent Cloud's medical industry model, focusing on the elderly services of Internet hospitals, to create a highly anthropomorphic virtual digital avatar "Xiaowei Nurse". The application enables hospitals to provide warmer and more convenient medical services for elderly patients, effectively improving the patient experience.
In the field of government affairs, large models need to be deeply applied in application scenarios to be more professional and reliable. Tencent Cloud's customized government model for tax departments breaks through the knowledge limitations of traditional intelligent customer service and significantly improves the problem interception rate and coverage. The new system intelligently handles the whole process from knowledge collection to operation, effectively reducing maintenance costs, and becoming a professional assistant in the tax field to keep up with the latest policies and regulations.
The in-depth application in multiple fields and scenarios benefits from Tencent's deep accumulation of full-stack large-scale model capability architecture. At the same time, Tencent Cloud is committed to continuously lowering the threshold for the implementation of AI large models and launching more plug-and-play products and services to meet the intelligent needs of different fields. In addition, Tencent Cloud is continuously improving the compliance and accuracy of its services, and optimizing its deployment model to better meet the growing intelligent needs of users.
In the future, Tencent Cloud will continue to export cutting-edge technologies and services to all walks of life, and work with ecosystem partners to provide continuous impetus for the intelligent and high-quality development of various industries.