The 2023 World Artificial Intelligence Conference opened at the Shanghai World Expo Center. The theme of this conference is "Intelligent Connected World Creates the Future". We will discuss and discuss the trends and laws of artificial intelligence development, opportunities and challenges, and work together to create a better intelligent world. . Xu Xiaolan, Vice Minister of the Ministry of Industry and Information Technology, Christian Bessiere, Chairman of the International Federation of Artificial Intelligence, Chen Jining, Secretary of the Shanghai Municipal Party Committee and other leaders delivered speeches.
As one of the important forums of this World Artificial Intelligence Conference, the World Artificial Intelligence Conference Large Model Open Source Construction Forum hosted by the Open Atom Open Source Foundation (hereinafter referred to as the "Foundation") was successfully held at the Shanghai World Expo Center on July 8. Sun Wenlong, chairman of the foundation, and Luo Qing, member of the Shanghai Economic Informatization Work Party Committee and leader of the Disciplinary Inspection and Supervision Team of the Municipal Disciplinary Commission in the Municipal Economic and Information Work Party Committee, attended the event and delivered speeches.
point one
The foundation vigorously promotes the establishment of the open source large model working committee
Sun Wenlong, Chairman of the Open Atom Open Source Foundation
Sun Wenlong said in his speech that at present, open source has become one of the most effective ways to build a technology ecosystem. Promoting the development of large-scale models in the form of open source and openness is conducive to gathering global wisdom and building a viable technology ecosystem that can be continuously iteratively updated. . Through the open source large model system, all enterprises can share data, share computing power, and jointly build algorithms, and various industries can also jointly create an ecosystem of diverse cooperation in technology, resources, and scenarios, thereby accelerating the iterative optimization of basic large models and promoting Large models apply multi-point practice.
Zhu Qigang, Director of Business Development Department of Open Atom Open Source Foundation
Zhu Qigang shared the "Introduction to the Work of Open Atom Open Source Large Models". He said that the open source innovation ecology of domestic large models still faces problems such as repeated resource investment, lack of coordination mechanisms, underlying resources and security guarantees. As a national foundation, the foundation has unique advantages in resource pooling and professional services. Aiming at the proposition of the era of open source construction of large models, it is planned to unite various professional institutions, integrate the strengths and resources of all parties, and build a neutral and advanced public service platform.
At present, the foundation is promoting the establishment of the open source large model working committee, aiming to rely on the open source public welfare platform of the foundation to organize forces from all walks of life, gather upstream and downstream partners, build a large model platform in an open source and open cooperation way, and provide computing power, data and The open source public welfare service of the algorithm accelerates the industrial application of large models and promotes the development of open source ecology of large models.
Aspect Two
Shanghai takes multiple measures to continue to promote the open source development of large models
Member of the Party Committee of Shanghai Economic Informatization Work
Leader of the Disciplinary Inspection and Supervision Team of the Municipal Commission for Discipline Inspection in the Municipal Economic and Information Work Party Committee
Luo Qing
In his speech, Luo Qing said that on the road to promoting innovation and development of large-scale models, China will adhere to open source collaboration, further promote the open sharing of computing resources and digital resources, and accelerate the formation of large-scale model industry ecology. Focusing on the development of large-scale models, Shanghai has actively taken measures to continuously establish cooperation and linkage mechanisms among governments, universities, scientific research institutions, industry organizations, and enterprises, increase algorithm open source, data openness, and computing power openness, and promote model co-creation, co-creation and sharing.
point three
Experts from production, learning, research, and use, jointly focus on the open source construction of large models
Wang Wei, Professor of School of Data Science and Engineering, East China Normal University
Wang Wei shared "Looking at the Development Trend of Global Large-scale Models from the Digital Insights of Open Source Communities", and introduced that open-source large language models have developed rapidly since 2018, with both activity and influence indicators increasing, and will tend to reach 2020 after 2020. Stablize. huggingface/transformers ranks first in terms of activity and influence, much higher than other models.
Zhou Wenmeng, technical director of Mota Community, senior technical expert of Alibaba
Zhou Wenmeng shared "United to the Future: ModelScope Magic Building Open Source Community x Large Model Ecology". He said that for ModelScope Agent, we have open sourced the corresponding Agent training tutorial. More interesting things.
Zhao Qiao, Head of Flying Paddle Product Team
Zhao Qiao shared "Increasing the model of the deep learning platform to promote the coordinated development of AI basic software and hardware", and introduced that the industrialization of large models still faces the challenges of high R&D threshold and difficulty, relying on the comprehensive support of algorithms, computing power and data. The development of the smart industry requires deep joint optimization of software and hardware.
Su Bin, General Manager of Beijing Huaheng Shengshi Technology Co., Ltd.
Su Bin shared ""Computing Power Scheduling" and "AI Large Model", and introduced in detail that the AI large model is the product of the combination of "big data + large computing power + strong algorithm". The training of the AI large model is a part of HPC. An important application, and the computing power scheduling system is an important core basic software of HPC, which plays a vital role.
Jiang He, Chief Scientist of the Key Laboratory of Industrial Big Data Integration and Application, Ministry of Industry and Information Technology
Jiang He shared "New Industrialization Service Platform - AI Digital Transformation Based on Modern Industrial Engineering". He said that in the data age, the two modernizations of industrial capabilities and industrial security are inseparable from the "data modeling infrastructure" and "trusted digital chain network infrastructure" of industrial enterprises. At present, both infrastructures have been deployed on the platform and will be open source soon.
Chen Ran, Founder & CEO of OpenCSG
Chen Ran shared "How to Build Core Competitiveness Based on Open Source Large Models". He said that in the open source ecosystem, building core competitiveness depends on the quality and richness of Dataset, the innovation and effect of Model, and the ease of use of Code. Important factors such as performance and scalability, as well as computing power optimization and distribution strategies.
Wang Wuyue, Director of AI Business at Shenzhen Data Exchange
Wang Wuyue shared "Large Model AI+: Shenzhen Data Exchange Empowers Large Model Vertical Industry Applications—Taking the Medical Industry as an Example". Middleware realizes the efficient allocation of data elements and computing power resources, and maximizes the industrial application value of data elements with the two-wheel drive method of "high-performance model base + high-quality vertical industry data".
Monkey Unlimited CEO Yin Bohao
Yin Bohao shared "Based on open source tool chains, truly realize open source large model tuning and application implementation". He said that model enhancement is the most worthy of attention. The process of large language model requirements to implementation is emerging, and the construction of large model workflows , which can unleash the full potential of large models.