Shulou(Shulou.com)06/02 Report--

Do you just want to be called a geek or do you want to be Bill Gates?

I think this is a multiple-choice question that every technologist can answer in less than a millisecond. But the key to this problem is that we need to know what the difference is between the two options.

Bill Gates can be said to be the most extreme and thorough person in the geek tide to industrialize technology. Perhaps it is not difficult to learn the rudiments of a technology and to develop it for fun. The hard part is to move into the industrial and commercial world and make your own development valuable and have a price.

Today's AI technology and developers are facing the same problem: the task of AI developers is not only to learn and explore, but to make their own development into the industrial world and truly be applied by production scenarios.

It can be said that the call of the times for AI to be implemented among specific practitioners is the transformation from AI developers to AI industrialists. But how do you accomplish this transformation? The evolution of technology tools and infrastructure is needed, and AI developers are also required to quickly learn the latest industry things and actively seize opportunities.

These two days, AICon Global Artificial Intelligence and Machine Learning Technology Conference has become a hot topic for a large number of AI developers. This year's AICon, the integration of AI and industrial applications has become the most important topic. During the conference, Huawei Cloud invited seven technical experts from different key fields of Huawei to share two aspects for the participating technology practitioners: the complete cycle of AI technology development and the industrial application of AI technology.

For Huawei AI system, Tan Tao, Huawei Shengteng AI engineer, shared the full-process application practice of Shengteng development tool chain and discussed the value of Huawei full-stack AI from the perspective of underlying development tools; Yu Fan, senior architect of Huawei MindSpore, analyzed the characteristics and application solutions of Huawei MindSpore from the perspective of AI computing framework; Wei Zhenqiang, senior solution architect of Huawei Cloud AI, discussed how AI developers can quickly complete AI application development based on full-stack capabilities from the perspective of AI development tools. In addition, Huawei senior technical expert Wei Junqiu shared Huawei's technical breakthroughs and application solutions for pre-trained language models; Huawei Cloud Senior Product Manager Zhao Yan shared how to build an efficient and trusted AI development environment based on Huawei Cloud IDE+Codehub.

At the level of AI technology and industrial application, Wang Xiaolei, an expert of Huawei Haisi Turing intelligent algorithm, shared the characteristics of Da Vinci architecture and rising chip, and explained the value and possibility of cloud edge full scene development for developers; Jia Yantao, chief technical expert of knowledge atlas of Huawei Central Software Institute, brought enterprise-level knowledge atlas construction, reasoning and application, helping developers understand the industrial application capability of Huawei full-stack AI.

Gathering seven dragon balls can summon dragon, Huawei cloud gathers experts from seven different industries at the same time, obviously also "the picture is very big." After sorting out the experts 'sharing, I found that Huawei Cloud's goal is to help AI developers sort out such a content: from basic development environment, to software and hardware combined with cloud edge collaboration, and then to real industry-level applications, how can AI development meet the transition from "geek hobby" to "industrial production" and maximize the efficiency of AI development for commercial applications.

And this process boils down to the chief keyword being "full stack."

We know that Huawei took the lead in proposing the concept of full-stack full-scene AI. For developers, the full stack is the primary condition for adapting to industry-level production, while the full scenario is the connection channel for AI development to trigger the application space.

What is the value of this AI foundation for developers? Today, we change our perspective and look at Huawei Cloud's AI system from the standpoint of AI developers.

Driven by the technical logic of full-stack full-scene AI, Huawei Cloud explores the possibility of becoming an AI industrialist for AI developers at three levels.

Full-stack AI tooling: the core guarantee for efficient low-threshold development

The so-called Full Stack was originally a name for an engineer in the IT industry who was competent for all work from the underlying architecture to the front-end application. Because of the difficulty of communication between different technical levels and the compatibility problems, full-stack engineers have become the most valuable presence in the enterprise-in fact, full-stack engineers are often the highest paid IT teams.

(Wei Junqiu Huawei Senior Technical Expert)

When AI arrived, Huawei extended the word to its own AI industry system. That is, under the application, Huawei is the first company in the industry to package from chips, chip enabling platforms, operator platforms, development frameworks, and then development platforms.

At the AICon site, five technology bulls from Huawei jointly created the "complete cycle of AI technology development" related content, which is actually telling AI developers what Huawei's full-stack AI means for developers.

We know that AI development towards real industrial applications is actually a very complex and refined project. Developers don't just have to come up with a "working" model, they have to make it meet demanding industry standards. This requires very complex work, which may involve hundreds of development tasks such as operator development, model development, application development, etc.

For developers, development tools are the premise for everything to start. Without a complete AI development tool chain, work becomes isolated and incompatible islands. Eventually, developers are exhausted in compatibility problems, and AI's industrial dream is reduced to a bubble.

(Yu Fan, Senior Architect of Huawei MindSpore)

Starting from the bottom chip, Huawei AI can get through every cycle of operators, frameworks and development platforms. This allows Huawei to build a full-process tool system from basic model training to hardware deployment. Huawei calls it Shengteng Development Tool Chain. This system covers the neural network software flow, operator comparison, model visualization series of work. With these basic tools, developers can truly achieve high-precision, low-cost compatibility development, so that efficiency and effectiveness have a basic guarantee.

At the next level, every AI developer must use a deep learning framework. One of the basic problems of the mainstream framework is compatibility with the development platform and chip environment on the one hand, and the other is that the mainstream framework in Europe and America is more born from the academic environment, and there are many inadaptability in the face of industrial applications.

Based on this, Huawei has built MindSpore development framework in the full-stack AI system. This framework is characterized by being fully grown in an industrial and development environment, with lowering development barriers as a top priority. MindSpore adopts a new programming paradigm to realize algorithms as code; a new execution mode to fully cope with the complexity and diversity of AI deployment; and a new collaboration mode to realize cloud edge on-demand deployment, which is more suitable for industrial needs.

(Wei Zhenqiang, Huawei Cloud AI Advanced Solution Architect)

The application of such a framework can help developers further lower the threshold and improve efficiency. The cloud doesn't stop there. At the next level, each developer must go through a cumbersome training process and reasoning deployment cycle. In order to minimize the workload of developers, Huawei Cloud launched ModelArts one-stop AI development platform and HiLens cloud collaborative AI application development platform. ModelArts has the industry's first model training efficiency, and can carry out intelligent standards and intelligent screening of data, so that AI development can be completed quickly and minimally. HiLens, on the other hand, can seamlessly open up end-cloud collaboration, giving full play to Huawei's full-stack advantages in the most important field of machine vision.

At this point, we can see that Huawei Cloud provides AI developers with full process empowerment from the bottom tool chain to the industry-level development framework and one-stop development platform, compressing the workload of each process to the lowest and adjusting the efficiency to the highest. In addition, Huawei Cloud also opens up a large number of AI capabilities and development assistance capabilities to developers. For example, Huawei's large number of pre-trained language model technologies are open to developers to improve their industry-level development capabilities;CloudIDE+CodeHub builds a secure and trusted cloud development environment, enabling developers to develop AI more securely and quickly.

(Zhao Yan, Senior Product Manager of Huawei Cloud)

The whole process provided by Huawei Cloud to AI developers can be referred to as the integration and tooling of full-stack AI technology. Developers can accurately utilize Huawei's full-stack AI technology advantages based on one tool, thus making development extremely simple, efficient and the threshold plummeted.

When developers get efficiency up to a certain threshold, the next question arises: how does AI go to industrialization?

Software and hardware full-scene collaboration: the basic channel for AI to move towards industrialization

The full-stack full-scenario AI proposed by Huawei is actually an industrial logic progressive relationship. Without full-stack AI capability, full-scene deployment cannot be discussed. When the full-stack AI is opened, the chip layer capability can be efficiently called by AI developers, developers can quickly develop models and applications on the full-stack channel, and Huawei's AI chip and product system deployed in the whole scenario can accurately play its role.

(Tan Tao, AI Engineer of Huawei Shengteng)

The core secret of full-stack AI development and full-scene AI deployment is a set of computing architecture: Da Vinci. We know that Leonardo da Vinci was probably one of the most versatile men in human history. Huawei's self-developed tensor computing architecture can also handle a variety of neural network tasks with maximum efficiency, achieving optimal computing power and lowest power consumption.

Relying on Da Vinci architecture, Huawei launched Shengteng 910 and Shengteng 310 chips to meet AI training and reasoning needs respectively, followed by edge side chips. These chips are integrated into different products and deployed to various scenarios at the edge of the cloud, which constitutes the full-scene AI channel provided by Huawei Cloud.

(Huawei HiSilicon Intelligent Algorithm Expert Wang Xiaolei)

This means that AI developers who use Huawei's cloud development system can seamlessly connect to the hardware matrix of the ascending ecosystem to complete the transition from software to hardware. In fact, many AI bulls have told me that AI development is actually very happy, but to implement the model to the development board and hardware, it means an unimaginable workload; any mistake must be reversed, the horror of hardware directly destroyed, and endless engineering capacity needs.

When Huawei Cloud gets through the AI deployment capability of the whole scene, these pains will disappear instantly, and AI development that could not enter the real world can also explore industrial applications.

Therefore, Da Vinci's secret is actually the AI provided by Huawei Cloud to bridge reality.

Enterprise Mission: The Ultimate Self-Examination of AI Industrialists

After Huawei Cloud provides a complete set of development tools and development foundation, and opens up the full scene connection with the hardware edge cloud, what should AI developers do?

The answer to this question may be the AI development value of Huawei Cloud, which is finally scored by the industrial ecology. At the AICon site, technical experts from Huawei explained how to build knowledge graph technology that can support enterprise-level applications based on Huawei's full-stack AI capabilities.

(Jia Yantao, Chief Technical Expert of Knowledge Atlas of Huawei Central Software Institute)

Knowledge Graph, a technology proposed by Google in 2012, is now widely used in the Internet, mobile Internet, information visualization and other fields. But for the average developer, it is far less familiar than voice, vision and other fields. Huawei expert Jia Yantao explained on the spot how "Xiaobai Developers" completed the development of enterprise-level tasks. It also interprets Huawei's knowledge graph technology and how to complete application deployment in operators, terminals, VR and other fields.

The enterprise-level knowledge graph may be seen as part of Huawei's cloud AI development environment. It points to the AI development work that ordinary developers can complete comprehensive goals, complex data logic, and high-precision models based on Huawei full-stack AI.

From starting development to completing enterprise-level tasks, full stack is a technical driver that can help developers shorten development processes, reduce development costs and improve development quality to the maximum extent.

Driven by the full stack, the definition of AI development is gradually changing. Who will be the first to drive the AI industrial revolution of this era-this should be one of the most exciting stories of the great era.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.