On March 20, 2025, the 3rd AI Definition Vehicle Forum 2025 was held as scheduled in Shanghai International Automobile City. As an annual event in the field of autonomous driving, the forum attracted many industry leaders, technical experts and representatives of automotive companies. Yang Guang, Director of Huawei's Ascend Autonomous Driving Industry Solution, shared Huawei's Ascend Ascend's innovative practices and future plans in the field of large-scale model technology and autonomous driving with the theme of "Building an Intelligent Foundation for a Win-Win Intelligent Driving Future". His speech not only revealed the underlying logic of technological breakthroughs, but also showed how Huawei's Ascend technology can be implemented to promote autonomous driving into the fast lane. The following is a speech report compiled for you by China Exportsemi:
1. Large-scale model technology: an accelerator for autonomous driving
Yang Guang pointed out at the beginning of his speech that large-scale model technology is becoming the core driving force in the field of autonomous driving. With the continuous upgrading of large-scale model technology, its application in autonomous driving and intelligent cockpit is becoming more and more extensive. Whether it is a large language model or a multimodal model, it is bringing a more anthropomorphic, safe and convenient experience to autonomous driving. Yang Guang emphasized that large-scale model technology not only improves the intelligent level of autonomous driving, but also promotes the commercialization of technology.
Taking Deepseek technology as an example, Yang Guang mentioned that Deepseek solves the problem of missing data in large model training by generating high-quality industry data through reinforcement learning and a small amount of data. At the same time, Deepseek's distillation technology accelerates the deployment of large models on the device side, enabling car companies to obtain high-quality large language models and multimodal models at a lower cost. This technological breakthrough not only lowers the technical threshold of car companies, but also greatly improves the speed of autonomous driving technology.
2. Huawei's Ascend Technology Breakthrough: From Underlying Computing Power to Autonomous Driving Solutions
As Huawei's key focus in the AI field, Ascend AI has built a complete software and hardware ecosystem since its release in 2018. Yang Guang introduced that the core goal of Ascend AI is to provide strong computing power support for domestic and foreign car companies and promote the innovation and application of autonomous driving technology.
In terms of underlying technologies, Ascend AI has built a high-performance operator library through a deeply open CAN layer, and developed the MindSpeed suite to support a variety of model algorithms. These technological breakthroughs not only improve the efficiency of computing power, but also provide more flexible technology options for car companies. Yang guang specifically mentioned that the driving SDK of ascend ai has become an important tool for car companies to develop autonomous driving models. With the driving SDK, car companies can quickly call high-performance operators and preferred model libraries, greatly shortening the development cycle.
In addition, Ascend AI also supports the training and inference of large language models and multimodal models through the MindSpeed distributed training acceleration suite. Yang Guang said that the kit is not only compatible with the industry's mainstream deep learning frameworks, but also deeply optimized for Ascend hardware, enabling OEMs to achieve efficient model training and inference on the Ascend platform.
Pictured: Yang Guang, Director of Huawei's Ascend Autonomous Driving Industry Solution Director
3. The key to the implementation of technology: open-source cooperation and tool chain support
In the process of technology implementation, the open-source ecosystem and tool chain support play a crucial role. In his speech, Yang Guang emphasized that Huawei Ascend actively embraces the open-source ecosystem and has carried out in-depth cooperation with multiple open-source communities.
For example, Ascend AI partnered with the Pytorch community to become its tenth highest-level member. This means that users of Ascend AI can enjoy native support from the Pytorch community, further improving development efficiency. In addition, Ascend AI has partnered with the OpenMMLab community to support the development of 2D and 3D perception models in autonomous driving scenarios. Yang Guang said that these collaborations not only reduce the cost of technology access for car companies, but also accelerate the innovative application of large-scale model technology in the field of autonomous driving.
In terms of the tool chain, Ascend AI provides full-process toolchain support, covering all aspects from operator development to model inference. According to Yang Guang, Ascend AI's MindStudio development tool can help car companies quickly develop high-performance operators and improve development efficiency through features such as memory detection and in-depth visual tuning. In addition, Ascend AI also provides precision debugging tools and performance tuning tools to help OEMs quickly locate problems and shorten debugging time during model development.
4. Commercialization Case: How Huawei's Ascend Empowers Car Companies
Technological breakthroughs and the improvement of tools will ultimately serve the commercialization. In his speech, Yang Guang shared several successful cases of Huawei's Ascend in the field of autonomous driving.
The first is the successful application of Huawei's ADS (Autonomous Driving Solution). Yang Guang introduced that Ascend AI supported Huawei ADS to build AI clusters with more than 10,000 cards, solving the problem of computing power bottlenecks. With the support of Ascend AI, Huawei's ADS model training efficiency has been improved by 2.3 times, and the training period has been shortened from weeks to days. In addition, Ascend AI uses a full-stack fault detection mode to reduce fault detection and recovery times and improve system stability.
In addition to Huawei's own success, Ascend AI has also launched in-depth cooperation with a number of car companies. Yang guang mentioned that the driving SDK and the preferred model library have become important tools for car companies to develop autonomous driving models. With the support of ascend ai, automakers can quickly commercialize autonomous driving technology. For example, an automaker has successfully developed an end-to-end autonomous driving model based on the driving SDK of ascend ai and commercialized it in a mass-produced vehicle.
5. Future Outlook: How Huawei Ascend will continue to promote the adoption of autonomous driving
At the end of his speech, Yang Guang looked forward to Huawei's future plans for Ascend in the field of autonomous driving. He said that ascend ai will continue to focus on the underlying ai basic software and hardware, and empower more car companies through the driving SDK and large model support. At the same time, Ascend AI will also actively promote the commercialization of autonomous driving technology from the laboratory to production vehicles.
Yang Guang called on more car companies to join the Ascend AI ecosystem and jointly promote the advancement of autonomous driving technology. He said that technology sharing and ecological cooperation are the key to achieving a win-win situation for the industry. Through an open technology ecosystem and in-depth industry cooperation, Huawei Ascend hopes to work with automakers to build the future of autonomous driving.
6. Conclusion: The future of autonomous driving is here
Yang Guang's speech not only demonstrated Huawei's Ascend technological breakthroughs in the field of autonomous driving, but also revealed the path to technology implementation and commercial application. From the underlying computing power to the open-source ecosystem, from tool chain support to commercialization cases, Huawei Ascend is promoting autonomous driving into the fast lane through all-round technological innovation and ecosystem cooperation.
With the continuous progress of large-scale model technology and the deepening of industry cooperation, the future of autonomous driving has come. Huawei's Ascend practice not only sets a benchmark for the industry, but also provides a reference path for more car companies to implement their technologies. It is foreseeable that with the promotion of Huawei's Ascend, autonomous driving technology will usher in a broader development space, and the future of travel will also undergo earth-shaking changes.
