Shanghai: Establish a World-Class Artificial Intelligence Industry Ecology by the End of 2025
Notice from the Shanghai Municipal People’s Government Office on the Distribution of the Implementation Plan for “Modelling Shanghai with AI”
To the District People’s Governments of All Regions, and to the Offices, Commissions, and Bureaus of the Municipal Government:
The “Implementation Plan for ‘Modelling Shanghai with AI'” is hereby distributed to you for your strict implementation upon approval by the Municipal Government.
Shanghai Municipal People’s Government Office
December 20, 2024
Implementation Plan for “Modelling Shanghai with AI”
In order to thoroughly implement the national strategic deployment on accelerating the development of “AI+”, and to implement the “Regulations on Promoting the Development of Artificial Intelligence Industry in Shanghai” and accelerate the construction of an artificial intelligence “highland” in Shanghai, as well as the creation of an artificial intelligence world-class industrial cluster, this implementation plan is formulated.
I. Main Objectives
We aim to build a solid foundation by promoting the development of artificial intelligence, focusing on areas such as smart computing clusters, corpus supply systems, virtual-real fusion training fields, and industry-specific large models. We will develop key areas of productivity tools in areas such as intelligent terminals, scientific intelligence, online new economy, autonomous driving, and embodied intelligence. We will focus on accelerating application enablement in key industries such as finance, manufacturing, education, healthcare, cultural tourism, and urban governance. By 2025, we aim to build a world-class artificial intelligence industrial ecosystem, strive to achieve a smart computing scale of over 100 EFLOPS, produce around 50 significant industry open corpus application成果, establish 3-5 large model innovation acceleration incubators, and build a number of upstream and downstream enabling centers and vertical model training fields.
II. Strengthening Basic Infrastructure Enablement
(1) Building a Large-Scale Autonomous Smart Computing Cluster: We will support the computing needs of artificial intelligence innovation applications in the city by building a self-controlled smart computing support base. We will accelerate the research and development of self-controlled smart computing chips such as general graphics processors, application-specific integrated circuits, and programmable gate arrays. We will promote the development of autonomous software such as distributed computing frameworks and parallel training frameworks. We will build an adaptation center for self-controlled smart computing hardware and software, promote testing and cluster validation of self-controlled smart computing chips. We will cultivate cloud service providers for smart computing, explore integrated service models that combine training and inference. We will optimize the city’s smart computing public service platform and enhance its ability to coordinate and schedule computing resources. We will also improve the supply of green electricity to reduce the electricity cost of various smart computing centers in the city.
(2) Establishing a Multi-Level Corpus Supply System: We will establish a general and specialized corpus library to support the development of basic large models and vertical applications. Focusing on the training needs of cutting-edge large models, we will promote the construction of basic large model training corpora. Focusing on industry demands from finance, manufacturing, education, healthcare, cultural tourism, urban governance, and other industries, we will create a series of industry open corpora and test datasets. We will build a corpus public service platform and construct tool chain platforms for corpus processing, production, and operation. We will cultivate a batch of out-of-the-box corpus service products. We will explore establishing a new corpus open sharing revenue distribution mechanism.
(3) Construction of Virtual and Real Fusion Ultra-Large Training Fields:依托龙头企业和科研机构，打造虚实融合的超大型实训场，建设支撑实训场的高性能计算集群、高精度三维建模和高质量训练数据集，创建与物理实体对应的高精度仿真环境和仿真训练系统，率先赋能具身智能、自动驾驶等大模型实训。
(4) Accelerating Innovation in the Industry Base Large Model System: We will accelerate the innovative fusion of general and vertical large models and build a series of industry base large models. We will support industry enterprises to strengthen their foundation in large model (L0) and open source ecology, and promote the application of reinforcement learning, instruction fine-tuning, and thinking chain technology. We will speed up the cultivation of industry dataset-integrated industry base large models (L1) and encourage the research and development of technologies such as knowledge distillation, pruning, quantization, and parameter sharing. We will focus on scenario application large models (L2) and smart body research and application, cultivate a batch of industry application developers and integratorss, encourage enterprises to build model as a service platforms, and promote the landing application of large models.
III. Accelerating the Creation of Key Productivity Tools
…
IV. Promoting Key Vertical Area Applications
…
V. Accelerating the Construction of Innovation Application Ecology
…
This notice concludes with several specific measures to promote innovation in artificial intelligence including building an ecological environment
