On April 24, on the eve of the Beijing Auto Show, at the "2024 TIME DAY Tencent Smart Mobility Technology Open Day", Tencent released the "Global Intelligence" solution for the automotive industry, covering five core scenarios, including automotive R&D, production, marketing, service, and enterprise collaboration. At the same time, Tencent released an upgrade plan in smart car cloud, smart cockpit, and going overseas.
Dowson Tang, Senior Executive Vice President of Tencent Group and CEO of Cloud & Smart Industry Business Group, said, "Tencent is committed to being a 'digital assistant' and 'ecological co-builder' for the upgrading of the automotive industry. On the one hand, Tencent continues to consolidate its core technical capabilities such as cloud and graph to build a solid foundation for the intelligent development of automobiles. On the other hand, we are also working with upstream and downstream partners in the industry chain to actively explore the innovative application of cutting-edge technologies such as AI large models in various scenarios of the automotive industry, so as to drive the development of 'new intelligence' in the automotive industry with AI." ”
Tang Daosheng, Senior Executive Vice President of Tencent Group and CEO of Cloud and Smart Industry Business Group
At present, Tencent has provided cloud services for more than 100 car companies and mobility technology companies, of which more than 30 have used Tencent's overseas cloud services. By the end of this year, Tencent's smart car products will be installed in more than 15 million vehicles.
The large-scale model "global intelligence" solution for the automotive industry covers the five core scenarios of "research, production, cooperation, sales and service".
At the conference, Tencent released the "Global Intelligence" solution for the automotive industry, which has been applied in five core scenarios: automotive R&D, production, marketing, service, and enterprise collaborative office. Zhong Xuedan, Vice President of Tencent Smart Mobility and Head of Tencent Smart Mobility, introduced that Tencent's automotive industry model is being applied to more than 10 automotive industry partners such as Changan, GAC, FAW Toyota, Dongfeng Lantu, and Yiche.
For example, in the R&D process, Tencent Cloud's AI code assistant can help software engineers write code, supplement code, diagnose, and test code, and improve R&D efficiency for autonomous driving. In a pilot for an enterprise customer, the AI code assistant helped improve its overall developer efficiency by 7%.
In the production process, large models can help optimize the process design of automobiles, and at the same time, the "transparent factory" based on digital twins and AI technology can intelligently analyze production data to help optimize production line layout and process flow. In terms of enterprise office, Tencent Lexiang AI assistant allows employees to quickly search in the complex enterprise database, and with the help of the induction ability of large models, they can obtain conclusions in minutes. Tencent Meeting's AI assistant has intelligent capabilities such as automatic summarization of meeting content and meeting management. In addition, the large model can also effectively improve the ability to identify financial risks in the automotive field, help car companies actively identify various black and gray industry organizations, and defend against malicious loan fraud.
In the marketing and sales process, with the help of Tencent's AIGC and multi-modal capabilities, car companies can efficiently respond to the complex and ever-changing new media marketing environment. For example, digital sapiens sell cars live broadcasts, and intelligently generate posters and videos and other publicity materials. In the service process, the new generation of Tencent Qidian intelligent customer service is trained and fine-tuned based on the large model of the automotive industry, which can provide more professional answers to difficult questions about car use and maintenance, and solve the problems of insufficient service experience and unfamiliar products of customer service personnel.
In the in-vehicle service scenario, Tencent's intelligent cockpit solution TAI 5.0 has launched a large cockpit model, which is based on the mixed element general model, and adds professional data fine-tuning and task training in the automotive industry, bringing users a human-like intelligent partner and answering high-level tasks more accurately and meticulously. According to Sun Jue, general manager of Tencent's smart travel intelligent cockpit, based on Tencent's leading APP Agent (intelligent agent) capabilities, under the mature process, it only takes one week to learn and proficiently use hundreds of applications or applets, and there is no need for API to achieve in-depth voice interaction with on-board applets and APPs, which can fully understand the user's intention, replace the user's intelligent operation of various complex applications, and improve service efficiency.
Li Bo, Vice President of Tencent Smart Mobility, said that Tencent adheres to a pragmatic attitude towards technological development and is committed to providing truly implementable and executable large-scale model solutions for the automotive industry. Tencent is willing to work with partners in the automotive industry to start with mature products and scenarios, and help car companies gradually cultivate and build their own large-scale model capabilities.
It provides a full-stack large-scale model capability base to help car companies innovate and improve efficiency
Tencent provides the automotive industry with a full-stack large model capability base from models, computing power, AI engineering platforms to AI applications, helping to open up the "last mile" of large model applications.
Tencent's full-stack capability matrix for the automotive industry model
In terms of models, Tencent Cloud's automotive industry model, based on the full-link self-developed hybrid model, adds massive professional data of the automotive industry to pre-training, fine-tuning vertical tasks in the automotive field, and reinforcement learning. In terms of performance, Tencent Cloud's large model for the automotive industry has achieved industry-leading results in terms of Chinese reading comprehension (COT), end-to-end Q&A, and automotive industry-related tasks.
At the computing base layer, Tencent provides high-performance, high-bandwidth, and low-latency computing base support for large models in the automotive industry. The new generation of HCC high-performance computing cluster, which supports 100,000 GPUs and is one of the most powerful large model computing clusters in China, can train large models with trillions of parameters in four days at the earliest, and can also be fully applied in scenarios such as autonomous driving and simulation. Tencent's self-developed Xingmai network brings the industry's highest 3.2 Tbit/s communication bandwidth to the new generation of clusters, enabling ultra-large computing power clusters to maintain excellent communication overhead ratio and throughput performance, and GPU utilization increased by 40%. At the same time, through Tencent's self-developed training and inference framework, car companies can save 50% of the computing cost when training hundreds of billions of large models.
At the AI engineering platform layer, Tencent provides a set of easy-to-use and practical tool chains that cover the entire life cycle of large models, reduce the difficulty of large model training and fine-tuning, and improve R&D efficiency. For example, Tencent Cloud's TI platform's industry large model fine-tuning solution can help model developers and algorithm engineers solve data processing problems in one stop, pre-set 140 datasets of different types of large model fine-tuning task scenarios, and build a complete pipeline covering data processing links such as data cleaning, prompt optimization, data filtering, and data augmentation, which can help customers quickly prepare high-quality fine-tuned data. Tencent Cloud's large model knowledge engine can comprehensively process complex enterprise documents, perform document parsing, reasoning, and generation more efficiently, and significantly improve the end-to-end Q&A effect. With the help of Tencent Cloud's large-scale model knowledge engine, the Q&A accuracy of an automaker's car assistant has increased from 35% to 84%.
The intelligent vehicle cloud and light map solutions have been upgraded to improve the efficiency and quality of autonomous driving mass production
Zhong Xiangping, Vice President of Tencent and President of Tencent Smart Transportation and Mobility, said that Tencent adheres to the strategy of "vehicle-cloud integration and cloud-based map", and has continuously expanded the scope of cooperation with the automotive industry and deepened its business depth. As autonomous driving enters the stage of large-scale mass production, Tencent Smart Car Cloud has undergone a new upgrade to meet new challenges such as business continuity and cost.
First of all, with millions of smart cars spread across the country, the demand for real-time vehicle-cloud interaction, data recycling, and large-scale high-frequency OTA has put forward higher requirements for business continuity. In terms of basic network, Tencent Cloud Networking Service relies on Tencent's global backbone network to help car companies establish service nodes around the world, providing more stable and high-quality services for local intelligent connected vehicles and users nearby, and providing multiple SLA guarantee levels to meet different business data transmission needs. In terms of zone cloud, Tencent is the first in the industry to open a dual-cloud zone in East China and North China, which can flexibly support remote disaster recovery and provide basic support for the business continuity of intelligent driving based on the configuration of remote active-active vehicles. In terms of security assurance, Tencent's industry-leading security products such as iOA zero trust security management, secure data lake, and Xcheck application security detection tool provide comprehensive security protection for the R&D, testing, and operation of autonomous driving, ensuring business continuity.
Second, Tencent Smart Vehicle Cloud provides a full range of cost reduction and efficiency improvement solutions, from direct cost reduction of basic resources, to comprehensive cost reduction through architecture and solution optimization, to efficiency improvement in AI R&D, data mining and other fields, helping customers achieve comprehensive cost reduction. Tencent's autonomous driving simulation system, TAD Sim, has launched a proprietary cloud version, relying on Tencent's compliant cloud services to solve the pain points of long construction cycle and high maintenance cost of privatized cloud simulation projects. TAD Sim can also provide a massive library of more than 1,000 realistic high-precision maps of different road features, as well as a library of more than 2,000 logical scenarios, providing customers with convenient, efficient and ready-to-use solutions.
At the same time, Tencent has also launched a light map solution, which improves the assisted driving experience through SD navigation maps or HD Air light and high-precision maps, and supports ordinary highway applications nationwide. For example, Tencent helped Yuanrong Qixing create the industry's first high-end intelligent driving mass production solution that uses only navigation map data, and has achieved stable operation in complex urban areas.
Liu Shuquan, vice president of Tencent Smart Mobility, said that Tencent has provided autonomous driving cloud services for NIO, Mercedes-Benz, Bosch, NVIDIA and other companies. On the same day, Tencent Cloud also announced that it had signed strategic cooperation agreements with WeRide, Black Sesame Intelligence, Zhixing Technology and other enterprises to jointly create leading intelligent driving solutions.
In addition, Tencent Auto's overseas solutions have also been newly upgraded, helping Chinese automakers efficiently deploy overseas digital infrastructure and flexibly access and enrich overseas local ecosystems through core capabilities such as globally integrated cloud services and overseas smart cockpit solutions. According to Shen Pei, general manager of Tencent's smart mobility strategy, Tencent Cloud helped GAC Aion quickly deploy digital infrastructure for Southeast Asia, and completed the mass production of the Internet of vehicles in the Thai market within three months. Tencent has also launched an overseas smart cockpit solution, which can help Chinese car companies quickly and cost-effectively build a localized cockpit service ecosystem based on the browser.
On the day of the event, Wu Xinzhou, global vice president of NVIDIA, Zhang Xiong, deputy general manager of GAC Aion, Chen Rui, deputy general manager of GAC Daisheng Technology Co., Ltd., Wang Jinwei, director of the new business department of FAW Toyota Motor Sales Co., Ltd., Lv Hanchao, R&D director of MBUX intelligent cockpit and Internet of Vehicles of Mercedes-Benz China R&D center, Han Xu, founder and CEO of WeRide, Wang Yongliang, deputy general manager of Wutong Auto Union, and members of the board of directors of Allianz Shihe Group, Jean-Marc Pailhol, Chief Officer of Global Strategic Partnerships, and other partner representatives shared their forward-looking insights on the development of automotive intelligence and the results of cooperation with Tencent. In the future, Tencent will continue to be a digital assistant for the automotive industry, open up innovation with ecosystem partners, and jointly promote the development of "new intelligence" in automobiles.