Shengsi MindSpore 2.3 is newly released | Shengsi Artificial Intelligence Framework Summit 2024 was successfully held

News on March 22, Beijing time, the artificial intelligence framework, as the root technology of software, has become the core force to accelerate the development of large artificial intelligence models and promote the development of industrial intelligence. The Shengsi Artificial Intelligence Framework Summit 2024 with the theme of "Rising for Wisdom, the Source of Innovation" was held today at the National Convention Center in Beijing. It aims to bring together the innovative forces of the AI ​​industry, promote continuous innovation of root technology, and jointly build open source artificial intelligence New ecology.

As a mainstream open source framework in China, MindSpore is committed to building a vibrant developer community and promoting the development of China's artificial intelligence industry. In the past two years, open source framework products such as Shengsi MindSpore have continued to grow. IDC research shows that Shengsi MindSpore has made remarkable achievements in ecological construction. This effort has helped promote the innovation and development of China's artificial intelligence technology, bringing new opportunities and challenges to the industry. The continuous development of MindSpore provides developers with more tools and resources, promotes technology exchange and sharing, and injects more vitality and innovation into the entire industry. In the future, MindSpore will continue to work hard to continuously improve the ecosystem,

##He You, academician of the Chinese Academy of Engineering and vice chairman of the China Artificial Intelligence Society, said in his speech: The China Artificial Intelligence Society has worked closely with Shengsi MindSpore, and has joined hands with Pengcheng Laboratory and Shengsi MindSpore to carry out academic research. The Fund 2.0 project accelerates the original academic innovation and publication of paper results based on Shengsi, giving full play to the original power of academic talents, combined with Shengsi's technical originality, and supporting the prosperity of the artificial intelligence industry with academic innovation.

Zhou Jun, President of Huawei ICT Marketing, said: The parallel computing capabilities, simple programming capabilities, and convenient deployment capabilities of the AI ​​framework have gradually become the key success factors for large model training. Shengsi will continue to innovate and explore, focus on root technology, and use a more open technology system to allow partners and developers to more flexibly realize large model innovation and accelerate the implementation of large model applications.

Jack Dongarra, Distinguished Professor of Computer Science at the University of Tennessee, shares insights into the evolution of computing and frameworks after winning the 2021 Turing Award. He emphasized the importance of open source in software development and believed that open source has many advantages and can provide a better software foundation for the future. Professor Dongla also pointed out that by developing and promoting the open source framework MindSpore, software performance can be effectively improved. This trend of open source frameworks has a positive impact on the entire industry, bringing more innovation and progress to the field of computer science. His views caused people to think deeply about the future development direction of computer technology

AI for Science new paradigm, generative AI empowers aerodynamic shape design

Tang Zhigong, academician of the Chinese Academy of Sciences and chairman of the Chinese Aerodynamics Society, said: Based on MindSpore, the generative aerodynamic design large model platform breaks the traditional design paradigm and shortens the design time from months to minutes. , to meet the conceptual design requirements. In the future, the platform will be expanded to aviation, aerospace, shipbuilding, high-speed rail, energy, automobile and other industries, leading the leap-forward development of industrial equipment design and manufacturing capabilities.

The generative aerodynamic design large model platform is independently developed and innovatively developed based on the Shengsi AI framework. In the model development stage, the MindSpore framework and the fluid dynamics suite MindSpore Flow provide a comprehensive scientific computing algorithm library and model universal interface to improve model development efficiency; in the model training stage, the MindSpore multi-dimensional parallel interface is used, based on the computing power support of Chengdu Intelligent Computing Center. Models and data can be efficiently expanded; in the model deployment phase, Shengsi large model suite is used to integrate professional knowledge into the aerodynamic design large model platform. By connecting non-AI tools such as large language models, aerodynamic shape design models, aerodynamic prediction models, and Fenglei software in series, the aerodynamic design large model platform can support a variety of aerodynamic shape design scenarios.

MindSpore 2.3 is newly released, making large model development and training simpler, more stable and more efficient

MindSpore actively explores cutting-edge technologies to support native and efficient training of large models. Through eight original parallel technologies such as multi-copy and multi-pipeline interleaving, the cluster linearity reaches 90% (less than 60% in the industry). Through whole-graph optimization and sinking execution, the computing power utilization reaches 55% (less than 60% in the industry). 40%); Aiming at the common problems of high cluster failure rate and long recovery time, deterministic CKPT technology achieves fault recovery in 20 minutes by compiling snapshots.

In terms of large model deployment, Shengsi realizes the unification of scripts, distributed strategies, and runtime through an architecture upgrade that integrates training and push. The inference deployment of Baichuan2-13B only takes 1 day. In terms of large model inference, the inference throughput has been increased by more than 2 times through LLM Serving; the model compression tool Golden Cudgel 2.0 has been upgraded to compress hundreds of billions of large models by ten times.

In order to lower the development threshold, Shengsi continues to upgrade the MindSpore TransFormers large model suite and provides the MindSpore One generative suite. The whole process is ready to use out of the box and the entire large model process can be completed in one week. development.

Innovate AI scientific computing paradigm and incubate basic large models in the scientific field. MindSpore has been exploring AI biocomputing for many years. It has collaborated with top scientific research institutions and partners to create an AI biocomputing suite, including more than 20 SOTA models such as protein structure prediction and generation, to accelerate innovation in related fields.

Facing long-term planning, MindSpore deeply cultivates its technology and continues to evolve to help the large-scale model industry come to fruition.

Four Actions to Empower Academics and Ecology

##Ding Cheng, Chairman of Shengsi MindSpore Open Source Community, was at the conference Announced Shengsi’s four actions to empower academics and ecology. The Academic Paper Fund 2.0, co-operated by Shengsi, China Artificial Intelligence Society and Pengcheng Laboratory, will join hands with more than 50 global AI scholars in the next three years to explore new academic peaks and achieve both quantity and quality improvement; Shengsi Development Board Application Innovation Action , based on the Orange Pie development board, provides systematic cases, tutorials and support to help developers quickly get started, get started quickly, and quickly create personalized applications; Shengsi accelerates the incubation of native large models, through incentives, special technical support and market alliances Promotion to support more partners from Shengsi migration and adaptation to native development; Shengsi open source community internship activities allow developers to truly grow through code practice.

Awarded to MindSpore Outstanding Contribution Mentor and Developer

As Gitee - the largest open source code hosting in China Platform, Open Source China selected MindSpore as "Gitee China's Best Open Source Contribution Project". According to the Gitee Index 2.0, MindSpore performed well in various indicators and became No. 1 in the overall category index under the Gitee-AI field classification. Ma Yue, chairman of Open Source China, said: The prosperity of the ecosystem is inseparable from the contribution of every project, and MindSpore is a treasure Gitee cherishes.

Huawei open-sourced the MindSpore AI framework in March 2020. After being open-sourced, it received positive responses from domestic and foreign developers, with tens of millions of visits, exceeding 6.87 millionDownload Installed and used, it ranks first among the tens of millions of code cloud open source projects. It serves more than 5,500 companies and cooperates with 360 universities. Since 2023, the number of top conference papers published based on the Shengsi framework has exceeded 1,200, ranking first in China and second in the world among all AI frameworks, and has become the most innovative and dynamic AI open source community in China.

#This conference selected 10 "MindSpore Outstanding Contribution Mentor" and 15 "MindSpore Outstanding Contribution Developers" in recognition of their outstanding contributions to the community.