"Xiaofu! Xiaofu!" - "Xiaofu", the Futian government smart assistant launched by the Shenzhen Futian District Government Service Data Management Bureau, can accurately understand the people's consultation intentions and transform the words of ordinary people into the language of government affairs, changing the traditional one-stop network communication do mode. "Xiaofu" is smart and cute, largely due to Huawei Cloud Pangu Government Affairs Large Model.
Recently, the same new thing happened in 8 domestic mines - coal miners do not need to go down the mine, but work on the ground. Not only the working environment is more comfortable, but also safety accidents are greatly reduced. Covering more than 1,000 subdivided scenarios under the business processes of mining, excavation, machinery, transportation, transportation, and washing of coal mines, reducing the work intensity of coal miners, Huawei Cloud Pangu Mine large model can handle it with ease.
Why are there fewer people queuing up to handle business at banks? Because the Huawei Cloud Pangu Financial Large Model can pre-train various bank operations, policies, and case documents, and automatically generate processes and operational guidance for counter staff based on customer questions, it has reduced the number of operations that originally required an average of 5 times to 1 time, the processing time was shortened by more than 5 minutes.
In railways, manufacturing, drug research and development and other fields, there are countless successful cases like this. It is the industry-specific large models that have brought about "wonderful" changes in our work and life.
From cloud to AI, what remains unchanged is the “black land”
On August 15, the "Interim Measures for the Management of Generative Artificial Intelligence Services" (hereinafter referred to as the "Measures") jointly issued by the Cyberspace Administration of China and other seven departments officially came into effect. The first batch of large models that have been registered and opened to the public through the "Measures" are listed. Also announced one after another. Intensive open services for large models show that the implementation of large models is entering a critical period.
From "being able to compose poetry and painting" to helping thousands of industries realize digital and intelligent transformation, the to B industrial Internet is the main battlefield for large-scale model applications. Currently, industry intelligence is accelerating, and this is the theme of this year's Huawei Connected Conference.
IDC analyzed in the "2022 China Large Model Development White Paper" that large models have begun to be deeply integrated with fields and industries, showing great potential in scenarios such as search, recommendation, intelligent interaction, AIGC, production process changes, and industrial efficiency improvement. This has once again verified that large models can not only be applied in technology companies, but also take steps towards all walks of life. On the basis of the general model, enterprises in various industries are using fine-tuning or prompt methods to add differentiated content between tasks, greatly improving the utilization of the model.
2023 is the first year of AIGC, and large models are undoubtedly an important engine. After the lively "Battle of Hundreds of Models", we should think more calmly about how large models can be deeply integrated with industries and application scenarios. In the future, from the perspective of the industry, it is necessary to strengthen the matching of large models with the needs of real scenarios and promote the large-scale implementation of large models; from the perspective of manufacturers, the layout of large models is becoming increasingly perfect, and the next step should be to focus on the breadth of industry empowerment and in-depth continuous exploration, while consolidating the construction of products and platforms based on large models, and promoting large model technology from the laboratory to diversified industry scenarios.
By absorbing massive amounts of knowledge, large models can adapt to multiple business scenarios, thereby significantly lowering the threshold for AI development and application, shortening the cycle from technology to application, and moving AI from workshop-style development and scenario-based customization to industrial development. , scenario-based optimization and large-scale solution of industry problems become possible based on large models. In the era of All Intelligence, Huawei's goal is very clear: to support each organization to use its own data to train its own large model, and to allow each industry to use its own professional knowledge to develop its own large industry model.
As an enabler, in the "Hundred Gardens" of AI and large models, Huawei Cloud will continue to cultivate the "black soil" through continuous innovation in cloud base, AI platform, development tools, etc.
Who is the creator of the "Meta-Ability Engine"
As we all know, large models are usually built with "money, computing power, and data." If an ordinary industry company wants to build a large general model by itself, it is neither necessary nor cost-effective. Directly quoting mature basic large models, and then adding users’ own industry experience and data to build exclusive industry or vertical models has become a necessary path for enterprises to upgrade intelligently.
However, enterprise private data resources are of extremely important value to enterprise operations. In order to ensure data security and privacy compliance, in addition to multi-level technical means such as federated learning and confidential computing, many large enterprises prefer Yu keeps data locally to ensure that "data does not leave the domain".
So, who will build a communication bridge between large models and vertical industry applications, and who is the indispensable "ferryman"?
IDC pointed out that in the future, large models will give rise to new industry and service application paradigms. Enterprises need to speed up the construction of a unified base for artificial intelligence, integrate expert knowledge maps, and create a "meta-capability engine" that can serve cross-scenarios or industries. Huawei Cloud is the best creator of "meta-capability engine".
From All Cloud to All Intelligence, although the cloud is still the same base, with the traction and support of AI and large models, the connotation of the cloud is constantly expanding and enriching. How the cloud supports the implementation of large models has become a top priority. Judging from the practice of many cloud service providers, in order to support the implementation of large models in industry enterprises, cloud service providers almost "teach everything to each other": first, provide the most basic computing power support; secondly, capable cloud service providers also It has developed its own general large model to facilitate users' reference, reference and citation; thirdly, it provides the basic resources or platforms necessary for the implementation of the large model; finally, it integrates tools and platforms to open up the breakpoints and help enterprises one-stop Deploy and apply large models efficiently.
At this year's Huawei Full Connection Conference, Huawei Cloud released the latest Huawei Cloud Stack 8.3, which provides 13 categories of 100+ cloud services. By enhancing the four core capabilities of Pangu large model, industrial Internet, data element circulation, and software development production line, Huawei Cloud Government and enterprise customers use the cloud in depth to accelerate intelligent upgrades. Especially in the field of large models, through Huawei Cloud Stack hybrid cloud, AI cloud services and key products of large models can be deployed in the enterprise's data center, helping enterprises to build exclusive large models in one stop.
Shang Haifeng, President of Huawei Cloud Stack
It’s not difficult to build industry-specific models
Pangu opened up the world, and Huawei Cloud's work in the field of large models is also innovative. In April 2021, Huawei Cloud officially released the Pangu large model, which at that time included the NLP large model and the CV large model. Since then, Pangu large models have been galloping on the road of industry and exclusive large models. Scientific computing large models, drug molecule large models, Pangu mine large models, meteorological large models, etc. have come out one after another. As of the release of Pangu Large Model 3.0 on July 7 this year, Huawei Cloud has formed a three-dimensional large model matrix, covering the 5+N+X three-layer architecture: L0 provides 5 basic models, including language large models and visual large models. , multi-modal large models, large prediction models, and large scientific computing models, providing a variety of skills that meet the needs of industry scenarios; L1 is based on "basic large models + industry knowledge" and provides N large industry models that adapt to industry characteristics, such as Government affairs, finance, manufacturing, drug molecules, mining, orbit, meteorology, etc.; L2 provides X scenario model services, focusing on a specific application scenario or specific business, and providing customers with "out-of-the-box" model services.
It can be said that the Huawei Cloud Pangu large model was born for the industry, that is, for the sake of the industry, to facilitate customers from thousands of industries to build their own exclusive large models based on L0 and L1. Through Pangu's pioneering layered decoupling architecture design, following the iterative upgrades of Pangu's basic large models, users' large industry models only need to be trained twice based on existing data sets, and the iteration of capabilities can be realized simultaneously, thus ensuring that users always Everyone can enjoy the latest capabilities of Pangu large models.
Huawei will continue to improve the capabilities of general large models and develop engineering tools so that customers and partners can use models more conveniently and efficiently. In the process of serving large government and enterprise customers for many years, Huawei Cloud Stack has always adhered to the concept of "making technology easy to use, making scenarios simple, and making experience replicable". It also follows this idea when using the Pangu large model in industry scenarios. .
Huawei Cloud has rich experience in implementing AI applications in the industry. Huawei Cloud AI has more than 1,000 projects in various industries. Based on an in-depth understanding of the industry and the core Know-How of the industry, Pangu Big Model has learned public data from more than ten industries such as finance, government affairs, meteorology, medical care, and health, with the data volume of each industry exceeding 50 billion tokens. In addition, Huawei Cloud has also worked with partners to create aPaaS for seven major industries: industry, heating, government affairs, coal mining, education, electric power, and highways, building the deepest industry accumulation for the Pangu model.
The powerful technology and industry experience of Pangu Large Model are combined with the new hybrid cloud version of Huawei Cloud Stack 8.3 to localize the Pangu Large Model capabilities of the public cloud and deploy them to the enterprise's data center. This not only ensures data sovereignty, but also makes the large model more accessible. Deployment is simpler, safer and more efficient. It is particularly worth mentioning that Huawei Cloud provides a variety of large model deployment modes. For industry users, the most ideal matching method is to pre-train on the public cloud, fine-tune on the local hybrid cloud, and deploy to the edge cloud for inference.
Pangu Big Model not only does difficult things, but also helps customers from thousands of industries solve the most difficult problems, making every industry, every company, and everyone's work more efficient and easier. Shang Haifeng, President of Huawei Cloud Stack, said: "We have always adhered to the AI for Industries strategy and continued to move forward on the road of deepening the industry. Large models should not be the patent of a few companies. We hope that based on the powerful capabilities and capabilities of Huawei Cloud Stack and Pangu Large Model, Experience will allow the AI industry to flourish, allowing each enterprise to have its own exclusive large model."
Building a cloud base for an intelligent world, AI reshapes thousands of industries
The new version of Huawei Cloud Stack 8.3 hybrid cloud strengthens the capabilities of Pangu large models and aims to help enterprises easily create one-stop exclusive large models. Through functional innovation and capability enhancement, Huawei Cloud Stack allows intelligence to be integrated into mainstream industry application development, deployment, governance and other scenarios in a "moisturizing and silent" way, transforming artificial experience into industry intelligence and achieving scale The implementation and continuous iterative optimization require not only the core capability of large models, but also strong support from other aspects.
For example, in terms of software development, Huawei Cloud Stack 8.3 has upgraded the CodeArts software development production line, providing more than 30 core service capabilities such as collaborative development, coding standards, distributed construction, and vulnerability checking. Haizhou Company builds an industry-leading industrial software R&D platform based on CodeArts provided by Huawei Cloud Stack, practices the IPD R&D model, ensures zero deviation in delivery through structured processes and contracted R&D, achieves efficient collaboration of thousands of people, and creates original technologies in large-scale industrial fields. source".
In terms of promoting industry applications, Huawei Cloud has launched an industrial Internet reference architecture based on Huawei Cloud Stack, allowing the construction of the Industrial Internet to follow rules. The reference architecture includes five core capabilities: industrial cloud-edge collaboration, industrial data collection, industrial data fusion, industrial intelligence hub, and industrial application development. It can be flexibly combined to meet the differentiated needs of complex scenarios and provide a full range of services from data collection to intelligent applications. link capabilities.
In terms of accelerating the circulation of data elements, Huawei Cloud has launched a data element circulation solution. Through the digital intelligence fusion, blockchain and privacy computing capabilities provided by Huawei Cloud Stack 8.3, enterprises can realize the development of high-quality data products such as algorithms, models and data sets. , making data available and invisible; exchanging data space through EDS allows data to be trusted for circulation and authorized operation, ensuring that data does not leave the domain.
To build the cloud base for the intelligent world, Huawei Cloud Stack provides time-tested practical solutions; allowing AI to reshape thousands of industries, "Huawei Cloud Stack + Pangu Big Model" provides a path for every enterprise to have its own large model. and tools.
past/issue/return/review
Huawei Cloud Stack deeply drives the intelligent upgrade of the financial industry