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

On August 16, MediaTek officially announced that it and Baidu jointly launched the hardware ecological co-creation program of Flying Propeller and Wen Xin Da Model, and the two sides will jointly promote the adaptation of MediaTek hardware platform with Flying Propeller and Wen Xin Da Model. At present, the two sides are working together to optimize the execution effect of the Wen Xin big model on the terminal device, which will realize the cooperative work of the big model in the terminal and the cloud, and bring users a breakthrough generative AI application experience.

As an advanced technology to promote the further development of AI, compared with simply deploying generative AI applications and services in the cloud, terminal-side deployment of spanning AI has obvious advantages in saving server costs, protecting user information security, improving real-time performance and realizing personalized user experience. The cooperation between the two sides will support the running of the Wenxin model on smart phones, cars, smart homes, the Internet of things and other terminal devices based on MediaTek chips, enabling terminal devices to provide a more secure, reliable and differentiated experience.

MediaTek and Baidu have worked together for a long time to complete the adaptation of Paddle Lite lightweight reasoning engine based on MediaTek NeuroPilot artificial intelligence general software platform.

NeuroPilot is a local end-to-side AI computing solution provided by MediaTek for developers. It provides a complete and powerful software solution for the MediaTek SoC platform with built-in heterogeneous computing units such as CPU, GPU and APU (independent AI processor). NeuroPilot enables developers to write at once and deploy AI applications based on neural network models on existing and future MediaTek hardware platforms and product lines including smartphones, cars, smart homes, and the Internet of things. AI developers can use mainstream TensorFlow, Pytorch, TF Lite, Caffe, Caffe2 and other custom third-party general AI frameworks to build neural network models and connect them to NeuroPilot systems, giving full play to the powerful computing power of MediaTek AI systems, significantly improving the efficiency of the model running on devices, and reducing power consumption and memory consumption.

Dr. Li Yanji, Deputy General Manager of MediaTek Wireless Communications Division, said, "Baidu Flying Propeller Deep Learning platform and Wenxin Model are in the leading position in the field of AI. Both of us will focus on the deployment of AI model on the terminal side to provide users with a more secure and reliable, more personalized and lower-cost end-to-side generative AI solution, giving terminal devices more powerful AI capabilities. MediaTek, together with Baidu Flying Paddle and Wen Xin Da Model, through AI end-cloud integration, will bring more exciting AI innovation opportunities and product experiences for developers and end users. "

Under the tide of generative artificial intelligence, large models are becoming an important driving force in the new round of scientific and technological revolution and industrial change. MediaTek will continue to join hands with Baidu Flying Propeller and Wenxin Big Model to accelerate the deployment of terminal-side generative AI, and jointly create a soft and hard artificial intelligence large model platform, so that science and technology can benefit the public and empower thousands of industries.

About MediaTek

MediaTek is a global fab-less semiconductor company, leading the market in mobile terminals, smart home applications, wireless connection technology and Internet of things products. About 2 billion built-in MediaTek chips are listed worldwide in a year. MediaTek strives to innovate and empower the market, providing high-performance, low-power mobile computing technology, advanced communication technologies, automotive solutions and multimedia capabilities for products such as 5G, smartphones, tablets, smart TVs, voice assistant devices, wearables and automotive electronics. MediaTek strives to make technology products more popular because we believe that technology can improve human lives and connect with the world, and everyone has the potential to use technology to create infinite possibilities (Everyday Genius).

On Baidu Flying Paddle and Wen Xin Da Model

PaddlePaddle is China's first open source, open and feature-rich industrial deep learning platform independently developed by Baidu. It is based on Baidu's years of deep learning technology research and business applications, and integrates core framework, basic model base, end-to-end development kit, rich tool components, learning and training community. Flying Propeller is the first in the industry to achieve a dynamic and dynamic framework design, taking into account the needs of scientific research and industry. it is in the international leading level in the development of convenient deep learning framework, large-scale distributed training, high-performance reasoning engine, industrial model base and so on. Flying paddle is an important part of Baidu's "chip-frame-model-application" stack technology layout, adapts and optimizes with chip layer depth, and efficiently supports the production and application of Wen Xin big model represented by Wen Xin Yiyan. At present, Flying Propeller has gathered more than 7.5 million developers, serving a wide range of finance, energy, manufacturing, transportation and other fields, ranking first in the market share of China's deep learning platform.

Baidu Wenxin model originates from industry and serves industry, and it is a big model of knowledge enhancement at industrial level. Through the integration and development of large model and deep learning framework, Baidu has created an AI base for independent innovation, which greatly reduces the threshold for AI development and application, meets the application needs in real scenes, and really gives full play to the industrial value of large-scale AI applications driven by large models. One of the major features of the Wen Xin big model is "knowledge enhancement", that is, the introduction of knowledge graph, the integration of data and knowledge, which improves the learning efficiency and interpretability. Since its birth in 2019, Wenxin ERNIE has made a number of technological breakthroughs in language understanding, text generation, cross-modal semantic understanding and other fields, and won more than ten world championships in open and authoritative semantic evaluation. In 2020, Wen Xin ERNIE was awarded the SAIL Award, the highest award of WAIC in the World artificial Intelligence Conference. In November 2022, Wenxin large model was further upgraded, including 11 new large models, building the largest industrial large model system in the industry, and through the upgrading of large model tools and platforms and product applications based on large model technology, such as Wen Xin, Wen Xin Baizhong, etc., further reduce the threshold of large model industrialization and let more enterprises and developers enter the new stage of AI application.

Thanks to the joint optimization of the flying oar deep learning platform and the Wen Xin big model, the effect, function and performance of the 3.5 version of the Wen Xin big model have been comprehensively improved, and the basic model upgrade, fine-tuning technology innovation, knowledge point enhancement, logical reasoning enhancement and so on have been realized. The effect of supporting Wen Xin is better, the efficiency is higher, and the application is wider. According to IDC's "AI large Model Technical capability Evaluation report, 2023" evaluation shows that Wenxin Big Model ranks first in comprehensive score, and both the "algorithm Model" and "Industry coverage" get unique full marks on the two core indicators.

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.