In Zhao Yi's "On Poetry", there is a sentence, "The eyes are full of vitality and transformation, and the craftsmen of heaven are ingenious and strive for innovation."
This poem is very suitable to describe the status quo of AI today. In the past few months, ChatGPT has been shining brightly around the world, attracting strong attention from all walks of life to AI, especially large-scale model technology. The importance of computing power and hardware and software infrastructure required by large-scale models has reached an unprecedented height.
Under the joint promotion of many driving forces, developers, industrial scenarios, and scientific research all need the assistance of AI technology and AI infrastructure that keep pace with the times.
In this case, more and more attention is paid to Ascend AI. Ascend AI, which integrates hardware base, basic software, and developer tools, is also advancing with the times. Not long ago, the first stop of the national tour of "Create the Future, Enjoy the Extraordinary" Ascend AI Developer Creation Day 2023 was successfully held in Shenyang. In this event, we can not only learn about the in-depth integration of Shengteng AI with industry-university-research applications in Northeast China, but also learn about its latest progress and new upgrades in 2023.
It’s spring time, let’s learn about the latest upgrade of AI requirements and the advancement of Ascend AI with the times. The integration of the two in this splendid today is just a picture of “the sky’s workers are ingenious and the days are striving for innovation”. .
Today's Needs for AI Infrastructure
With AI truly deepening into the industry, based on scientific research, and the rapid rise of new AI technologies represented by large models, the demand for AI technology, especially AI training and deployment-related infrastructure, has ushered in a comprehensive upgrade. At the time point of 2023, we can see the following characteristics of AI infrastructure in different fields:
1. The industry needs to be real: Nowadays, the path to explore intelligence in various industries is very clear. The combination of AI and industries has begun to migrate from the exploration stage to the practical stage. Therefore, all industries need to be able to be deeply integrated into the industry, and put forward practical, scene-oriented, and industrial-oriented requirements for related products and software.
2. Academic needs: The combination of scientific research and AI technology is in the full blossoming and vigorous exploration stage. The academic community needs large-scale AI devices for different disciplines and different research fields to support and leverage the new paradigm of scientific research + AI.
3. Developers need to be simple: For developers, since AI technology has become mature, it is necessary for AI infrastructure to be easier to use and easy to get started, and integrate abundant and mature development tools and development examples to minimize the development threshold.
Facing these new demands, AI software and hardware basic testing is the foundation and foundation. If the foundation is not strong, the earth will shake and the mountains will shake. The reason why Ascend AI is valued is precisely because it can become a general AI foundation for industry, academia, research and developers. At the same time, it also has the ability to keep pace with the times, responding and meeting these latest needs with the highest efficiency.
Smart help for industry
Today, the Ascend AI basic software and hardware platform has developed into a full-scale platform including the Atlas series hardware, the heterogeneous computing architecture CANN, the full-scenario AI framework MindSpore, the Ascend application enablement MindX, and the one-stop development platform ModelArts. Process AI infrastructure can help partners and developers efficiently use AI capabilities and innovate scenario-based AI applications.
At the level of industrial ecology, Ascend AI adheres to the strategy of open hardware, open source software, enabling partners, and developing talents, and helps the intelligent development of the industry through comprehensive and diversified methods, enabling the AI industry to land.
At the hardware level, we can see that Ascend AI has been able to deeply integrate with industry scenarios and industry needs, bringing AI computing power to actual productivity applications. On the day of "Creation and Sharing Day", China Coal Science Research Institute and China Coal Science Robotics Co., Ltd. released a robot solution based on Shengteng AI for industry developers and partners. 500 intelligent small stations are applied to robot image recognition, so that intelligent inspection robots can analyze potential safety hazards. This solution has been applied in substations and coal mine scenarios, promoting the intelligent upgrading of the coal mining industry.
At the software level, Ascend AI can empower the industry through frameworks, development tools and other levels, combined with abundant AI computing power, to meet the software requirements of AI models from training to deployment. Especially when large models become a new demand for industry intelligence, the value of Ascend AI's native support for large model training and industrial implementation is highlighted.
The cooperation between Northeastern University, Huawei, and Shenyang Artificial Intelligence Computing Center has officially entered a new stage. The first large-scale model in Liaoning based on Ascend AI - "Shenyang. The One" (The One) multimodal recommendation model was officially released. The data parameters of the "Shenyang.Taiyi" model have reached 1 billion, which can integrate multi-modal information such as text, images, social networking, and knowledge graphs, analyze massive data, and recommend accurate content. This technology has matured and will soon bring great value to the development of big data and Internet industries in Northeast China.
In addition, we can also see Shenyang Meixing Technology Co., Ltd. based on Shengteng AI, incubating the "Wisdom · Joyful" road data reconstruction large-scale model on the Creation Day, and Liaoning Liaohe Laboratory and Huawei, Shenyang Shengteng The artificial intelligence ecological innovation center signed a contract to formally start the comprehensive cooperation in the field of artificial intelligence of Shengteng. The artificial intelligence pre-training model for defect detection developed by Liaoning Liaohe Laboratory based on Ascend AI technology has completed the first phase of development, and the three parties will start construction around the "Future Industrial Internet Research Infrastructure" platform.
The foundation of scientific research and innovation
The integration of scientific research and AI technology has been widely expected by the academic community. It is likely to change the scientific research paradigm of multiple disciplines and comprehensively upgrade scientific research efficiency. Such profound changes also require a comprehensive upgrade of AI infrastructure to cooperate.
To this end, Ascend AI has created a brand-new AI+scientific computing fusion architecture to meet the new computing paradigm of deep integration of research methods such as supercomputing, high-performance computing, and AI computing in scientific research scenarios, changing the inability of different computing solutions in scientific research to be integrated and isolated from each other. The situation has built an innovative foundation that is more in line with the real needs of scientific research + AI.
Not only that, Ascend AI also aims at the integration of large-scale models and scientific research, launched the Ascend Scientific Research Innovation Enablement Plan, promoted the implementation of large-scale model innovation in scientific research, and created a series of examples and examples of the combination of large-scale models and scientific research scenarios.
For example, as we have discussed before, based on the Ascend AI software and hardware base, an efficient and high-precision AI simulation and prediction large-scale model "Dongfang. Yufeng" for the airfoil flow field of large passenger aircraft has been created. With the support of Shengsi MindSpore fluid simulation suite, it can greatly improve the efficiency of fluid mechanics research and greatly reduce the simulation time.
Based on Ascend AI's "Pengcheng. Shennong" platform, it can simulate the mutation of virus strains including the new coronavirus in a very short period of time, so as to predict the direction of virus mutation in advance, improve scientific research capabilities, and strengthen the public health field. Ability to handle in advance.
With the help of AI+scientific computing fusion architecture, scientific research can accommodate AI technology and large-scale model innovation with higher efficiency and low threshold. This is especially important in basic science and disciplines related to national economy and people's livelihood. The construction of laboratories and new scientific research bases provides new infrastructure options.
Convenient choice for developers
Back to the developer level, AI hardware development products that can be used quickly and easily applied are currently scarce resources on the market. Developers often fall into problems such as no product available, or find that the development cost is too high and the learning cost is too high after obtaining the product. It is the most simple and direct way to cut into the needs of developers and provide development resources that effectively connect industry scenarios and business value. Mission-critical tasks in the new phase of AI infrastructure.
To this end, Ascend AI will launch the Atlas 200I DK A2 developer kit. The new product is oriented to developers, provides an out-of-the-box development experience, and integrates rich reference examples, so that developers can not only obtain infrastructure, but also learn and understand the latest development technology and development experience.
From the perspective of simplicity, the Atlas 200I DK A2 developer kit can be used to build a development environment in half an hour and complete the first inference application development in 2 hours, realizing real out-of-the-box development.
In terms of learning examples, it can provide three end-to-end scenario-based demos, including smart cars, smart robotic arms, and smart voice interaction, which are the most mainstream and most widely used scenarios in the real industry. Debug your own algorithms in a highly simulated environment, without having to reinvent the wheel, and devote yourself to high-value development work as soon as possible. In addition, the development kit is also equipped with more than 100 scene samples, including view analysis, OCR, action/gesture recognition and other fields.
In terms of community resources, the Ascend community can provide developers with hundreds of high-performance pre-training models, covering all mainstream and popular AI directions. Reduced the difficulty of model migration to Ascend AI.
In addition, the Atlas 200I DK A2 developer kit can also provide full-process development tools and a wealth of available interfaces, allowing developers to be in real and usable development scenarios and experience convenient and minimal development efficiency.
With the help of similar new products, Ascend AI continues to narrow the distance with developers, directly meets the needs of developers, innovates the multi-level development assistance capabilities of software and hardware, and enables developers to succeed.
Up to now, Ascend AI has developed more than 1.3 million developers, opened courses related to Ascend AI in more than 200 colleges and universities, and joined hands with more than 1,000 industry partners to jointly incubate more than 2,000 solutions.
To sum up, whether it is to promote AI computing power and large-scale models to meet the real needs of the industry, to build a new intelligent computing paradigm for scientific research, or to provide developers with a minimalist and efficient development base, Shengteng AI has achieved great success in multiple industries. Dimensions is innovating with the times.
Ascend AI listens to the industry, understands scientific research, understands the attitude of developers, and completes the mutual growth of AI demand and AI infrastructure supply and demand, and supports each other.
If we say that the global explosion of large models and the sailing of digital China constitute the "Tiangong" of AI technology, then the cooperation between Shengteng AI and partners and developers has become a concrete and real "Tiangong" The "human ingenuity" of humanization.
Only by combining the two can we have a new atmosphere and a new era in which AI technology is competing day by day.
