ON AUGUST 16, THE WAVE SUMMIT DEEP LEARNING DEVELOPER CONFERENCE 2023, HOSTED BY THE NATIONAL ENGINEERING RESEARCH CENTER FOR DEEP LEARNING TECHNOLOGY AND APPLICATION, WAS HELD. Wang Haifeng, Chief Technology Officer of Baidu and Director of the National Engineering Research Center for Deep Learning Technology and Application, elaborated on the four capabilities of big language models to understand, generate, logic and memory, with the title of "Big Language Model Brings Dawn to General Artificial Intelligence", which brings the dawn to general artificial intelligence. At the meeting, Baidu released a series of technologies, products and ecological achievements such as Wenxin big model, flying propeller platform, and AI native application such as flow. Thanks to the mutual promotion of the industry-level deep learning open source platform and the Wenxin model, the flyblade ecology has become more prosperous, and it has gathered 8 million developers, served 220,000 enterprises and institutions, and created 800,000 models based on the flypaddle.
Baidu CTO Wang Haifeng keynote speech
At the meeting, Wang Haifeng explained the timeless meaning of the Chinese name "Galaxy Community" of AI Studio, a developer community of flying oars, "Wen Xin plus flying oars, go to the galaxy". Together with all developers, with the blessing of Flying Oars and Wen Xin, we will build a Galaxy community and go to the sea of stars of general artificial intelligence.
Wang Haifeng further said that artificial intelligence has a variety of typical capabilities, understanding, generation, logic, memory is the core basic capabilities, the stronger these four capabilities, the closer to general artificial intelligence, and the big language model has these four capabilities, bringing the dawn of general artificial intelligence.
Taking Wen Xin Yiyan as an example, as a knowledge enhancement large language model independently developed by Baidu, Wen Xin Yiyan first fuses learning from trillions of data and hundreds of billions of knowledge to obtain a pre-trained large model, and adopts technologies such as supervised fine tuning and human feedback reinforcement learning and prompting, which has technical advantages such as knowledge enhancement, retrieval enhancement and dialogue enhancement.
In recent months, Wen Xin's words have continued to evolve rapidly. Wu Tian, vice president of Baidu Group and deputy director of the National Engineering Research Center for Deep Learning Technology and Application, released five native plug-ins of Wen Xin Yiyan: Baidu Search, Scroll Document (document-based interaction), E Yan Yi Tu (data insight chart generation), Say Picture Painting (image-based interaction), One Mirror Flow Shadow (text to video), Wen Xin Yiyan now supports the use of 3 plug-ins at the same time, which not only expands the capability boundary of large models, but also better adapts to the needs of scenarios.
Wu Tian, Vice President of Baidu Group and Deputy Director of the National Engineering Research Center for Deep Learning Technology and Application, shared the latest release of Wen Xin
Wu Tian revealed that AI Studio (Galaxy Community) is already the largest AI community in China, condensing 6.09 million development projects. With the advancement of big model technology, big model developers are also rapidly gathering in this community. At the conference, AI Studio (Galaxy Community) carried out the latest upgrade and officially launched the Galaxy Big Model Community, allowing developers to get an integrated large model development experience. Flying Paddle + Wen Xin runs through the whole development process and continues to co-create large model projects and applications with developers. At present, the Galaxy Big Model Community has accumulated more than 300 large model creative applications, and the community also provides a wealth of functions to facilitate developers to communicate. At the same time, the invitation test of the large model plug-in was officially launched, and the large model plug-in relies on the core technology of "Wen Xin Yiyan" to provide you with a plug-in development toolset, which can carry out information services, tools, and plug-in development based on large language model innovation. After being connected through the platform, it can be closely integrated with the application layer ecosystem. Wu Tian also announced the launch of the Wenxin model "Galaxy "Galaxy Co-creation Program" to provide ecological and industrial resources and jointly promote AI application innovation with developers.
The rapid progress of Wen Xin Yiyan benefits from the mutual promotion of the flyblade industrial-level deep learning open source platform and the Wen Xin big model. It is reported that in terms of efficiency, through the end-to-end adaptive hybrid parallel training technology of flying paddle and collaborative optimization of compression, inference, and service deployment, the training speed of Wenxin large model has reached 3 times and the inference speed has reached more than 30 times.
At the conference, Ma Yanjun, general manager of Baidu's AI technology ecosystem, officially released version v2.5 of the open source framework of Flypad, and the infrastructure was fully upgraded. The new architecture has improved differentiation capabilities and more versatile performance optimization. At the same time, the flying propeller scientific calculation tool component PaddleScience v1.0 was officially released, and the flying propeller PaddleHelix released the HelixDock protein-compound conformational affinity prediction large model.
Ma Yanjun, General Manager of Baidu AI Technology Ecology, shared the latest release of the flying paddle platform
Flying paddle has also promoted the construction of national standards for software and hardware adaptation. Led by the China Electronics Standardization Institute, Baidu, Sugon, Feiteng and Inspur jointly drafted the national standard "Artificial Intelligence Deep Learning Framework Multi-hardware Platform Adaptation Technical Specification". Based on this national standard, Wenxin Big Model has been adapted with 12 hardware partners such as NVIDIA, Zhongke Shuguang, Kunlun Chip, Cambrian, Huawei, Tianzhixin, Bicheng Technology, Qualcomm, Intel, MediaTek, Moore Thread, and Arm Technology, covering a variety of hardware types on the cloud and device side.
At present, artificial intelligence represented by large language models is penetrating into thousands of industries, accelerating industrial upgrading and economic growth. Similarly, it also brings unprecedented development opportunities for the corporate office field.
Based on AI native thinking, Li Ying, Vice President of Baidu Group and Chief Information Officer of Baidu Group, announced that Baidu's intelligent work knowledge management concept "Innovation Pipeline = AI X Knowledge Management" has been upgraded to 2.0. At the same time, Baidu Ruliu "Super Assistant" was officially released. At the scene, Li Ying gave more than 10 demonstrations to the "super assistant" around four scenarios such as mobile task execution, intelligent document processing, CBI conversational business intelligence, and intelligent communication.
Li Ying, Vice President of Baidu Group and Chief Information Officer of Baidu Group, shared the practice of AI native applications
For example, "voice one-click travel", on-site voice input "I go to the Shanghai R&D center for a meeting the morning after tomorrow, return to Beijing on the same day, please help me plan my trip", the super assistant can quickly call the functions of different modules or platforms, automatically submit travel applications, book round-trip air tickets and hotels, etc., just like the real assistant, the efficiency is compressed from hour-level operations to minutes-level or even second-level completion.
In the research and development of intelligent scenarios, Baidu Comate series products have been newly upgraded, and the intelligent programming assistant Comate X and Comate Stack tool suite have been released. Comate X intelligent programming assistant not only provides basic code intelligent recommendations, but also helps engineers "think", "write" and "change" in the whole life cycle and scenarios of R&D. In addition to code explanation, code generation, etc., it also has interline comment generation, unit test generation, document generation, command line generation, interface code generation and other capabilities, and currently supports more than 30 programming languages and more than ten IDEs. At this conference, as the first commercial all-scenario intelligent programming assistant in China, Comate X was officially opened to enterprise developers. Comate series products have been fully used within Baidu, and have served and docked with more than 100 enterprise partners, and are in a mature commercialization stage.
The major changes in the enterprise office field are a microcosm of the artificial intelligence represented by the big language model to accelerate the transformation and upgrading of all walks of life. In this process, technological innovation and application landing form a virtuous circle, understanding, generation, logic, memory and other capabilities continue to improve, the breadth and depth of industrial applications continue to expand, and the big language model brings the dawn of general artificial intelligence.