Big data industry innovation service media
——Focus on data and change business
At the Wave Summit 2023 Deep Learning Developers Conference held on August 16 , Wang Haifeng , chief technology officer of Baidu and director of the National Engineering Research Center for Deep Learning Technology and Application, said that the big language model has four core foundations of understanding, generation, logic, and memory. ability, bringing the dawn of general artificial intelligence.
Flying paddle ecology brings together 800 developers and creates 800,000 models
In Wang Haifeng's view, the typical abilities of artificial intelligence, such as creation, programming, problem solving, planning, etc., all depend on the four core basic abilities of understanding, generation, logic, and memory to varying degrees. Taking problem solving as an example, from understanding the problem, to answering the problem, to finally writing the answer, artificial intelligence is nothing more than a combination of the four major abilities of understanding, memory, logic, and generation.
How to obtain the four core basic capabilities of artificial intelligence? The first is to integrate learning from trillions of data and hundreds of billions of knowledge to develop a pre-trained large model. On this basis, intensive learning is realized through fine-tuning and feedback, and finally the large model is enhanced with knowledge, retrieval, and dialogue. . Deep learning technology is the only way to general artificial intelligence.
In the exploration of deep learning framework technology, Baidu launched an open source and open flying paddle deep learning platform . Flying paddle deep learning platform + Wenxin large model , mutually promote and promote each other, constitute the base of Baidu's artificial intelligence strategy.
According to Wang Haifeng, after years of development, Baidu Flying Paddle Ecology has gathered 8 million developers, serving 220,000 enterprises and institutions, and created a total of 800,000 models based on Flying Paddle.
At the Wave Summit, Baidu officially introduced the Chinese name " Xinghe Community " for AI Studio, a flying paddle developer community. Wang Haifeng said that Baidu is willing to work with developers to jointly build the Xinghe community and go to the stars and oceans of general artificial intelligence with the support of the Flying Paddle deep learning platform + Wenxin large model.
A technological breakthrough in the framework layer, a comprehensive upgrade of the open source framework of Flying Paddle
In the era of artificial intelligence, the IT technology stack has undergone fundamental changes, from a chip-operating system-application three-tier architecture to a chip-framework-model-application four-tier architecture. Among them, the deep learning framework can be called the operating system in the era of artificial intelligence. It connects to various chips and cloud computing platforms downwards, and undertakes various AI business models and industry applications upwards.
In the national "14th Five-Year Plan", the deep learning framework is included in the category of "new generation artificial intelligence" and has become a cutting-edge technology that is mainly supported.
In 2010, Baidu announced All in AI, a full-stack layout of artificial intelligence. After investing hundreds of billions of dollars in research and development, Baidu has gradually made breakthroughs in the field of artificial intelligence: at the chip layer, there is Kunlun Core; at the framework layer, there is the Flying Paddle deep learning platform; Layers include intelligent search and automatic driving.
The Flying Paddle deep learning platform, which was open-sourced in 2016, is a major breakthrough in Baidu's artificial intelligence framework layer technology. It has become one of the preferred platforms for domestic artificial intelligence developers by competing with two foreign deep learning frameworks, TensorFlow and PyTorch.
At the Wave Summit conference, Ma Yanjun, General Manager of Baidu AI Technology Ecology, officially released the v2.5 version of the open source framework of Flying Paddle. At the same time, PaddleScience v1.0, a scientific computing tool component of Paddle, was officially released, and PaddleHelix, a large model for predicting protein-compound conformational affinity of HelixDock, was released .
Galaxy Community, the largest AI community in China, with a collection of 6 million projects
After opening up the Flying Paddle deep learning platform and Wenxin large-scale model, Baidu is working hard to build a Flying Paddle + Wenxin ecology, with software and hardware adaptation and Galaxy community building becoming the focus of development.
In terms of software and hardware adaptation, led by the China Institute of Electronic Technology Standardization, Baidu, Sugon, Feiteng, and Inspur jointly drafted the national standard "Technical Specifications for Adaptation of Artificial Intelligence Deep Learning Framework to Multiple Hardware Platforms". Based on this national standard, Baidu Flying Paddle, Wenxin Large Model and Nvidia, Sugon, Kunlun Core, Cambrian, Huawei, Tianshu Zhixin, Biren Technology, Qualcomm, Intel, MediaTek, Moore Thread, Amou Technology 12 other hardware partners have carried out adaptation, covering cloud and end-side multiple hardware types.
In terms of community, AI Studio (Galaxy Community) is already the largest AI community in China, bringing together 6.09 million development projects. With the advancement of large-scale model technology, large-scale model developers are also rapidly gathering in the Galaxy community. At the Wave Summit, Wu Tian, vice president of Baidu Group and deputy director of the National Engineering Research Center for Deep Learning Technology and Application, announced the latest upgrade of the Galaxy community and officially launched the Galaxy large model community. According to reports, the Xinghe large model community has accumulated more than 300 creative applications of large models.
At the conference, Wu Tian announced the Wenxin large-scale model Galaxy Co-creation Plan, Baidu provides ecological and industrial resources, and jointly promotes artificial intelligence application innovation with developers.
Flying paddle + Wenxin fly together, Wenxin releases five major plug-ins
On March 16, Baidu officially released Wenxin Yiyan, a knowledge-enhanced large-scale model. In less than half a year, Wenxinyiyan, which accelerates evolution and iteration, has launched version 3.5.
The rapid progress of Wenxinyiyan has benefited from the mutual promotion of the open-source open platform of Flying Paddle's industry-level deep learning and Wenxin's large model . According to Wu Tian, in terms of efficiency, through the end-to-end self-adaptive hybrid parallel training technology and the collaborative optimization of compression, inference, and service deployment, the Wenxin large model training speed has reached 3 times the original speed, and the inference speed has reached more than 30 times.
At the Wave Summit conference, Wu Tian released five native plug-ins of Wenxin Yiyan: Baidu Search, Scroll Documents (document-based interaction), Eyan Yitu (data insight chart generation), Shuo Diagram ( Image-based interaction), Yi Jing Liu Ying (text-to-video). Wenxinyiyan supports the use of three plug-ins at the same time, which not only expands the capability boundary of the large model, but also better adapts to the needs of the scene.
With the help of the plug-in, Wenyan Yiyan demonstrated the black technology of producing commercials in seconds and completing codes in a few words .
Baidu Chief Technology Officer Wang Haifeng said that in the future, Baidu will work with developers to build a plug-in ecosystem and share technological innovations. Grasping technology + ecology with both hands, the application of Flying Paddle + Wenxin has landed and is running out of acceleration.
Text: Bugle-X / Data Ape
