AI Reshapes Qianxing Baiye Huawei Cloud Releases Pangu Big Model 3.0 and Shengteng AI Cloud Services

On July 7, 2023, the Huawei Developer Conference (Cloud) officially opened in Dongguan, China. It also has branch venues in more than 10 countries worldwide and more than 30 cities in China, inviting global developers to gather together to exchange and share new industrial opportunities and technological practices under the AI wave

On July 7, 2023, the Huawei Developer Conference (Cloud) officially opened in Dongguan, China. It also has branch venues in more than 10 countries worldwide and more than 30 cities in China, inviting global developers to gather together to exchange and share new industrial opportunities and technological practices under the AI wave.

In the keynote speech of the conference held on the afternoon of the 7th, Huawei Executive Director and Huawei Cloud CEO Zhang Ping'an launched the Pangu Big Model 3.0 and the Ascension AI Cloud Service. Among them, Pangu Model 3.0 revolves around the three innovative directions of "industry reshaping", "technology rooting", and "open and flying", continuously building core competitiveness and providing better services for industry customers, partners, and developers. The Shengteng AI cloud service provides a single cluster with 2000PFlops of computing power, with a 30 day long stability rate of 90% for thousand card training, providing stable and reliable AI computing power for the industry, making big models within reach.

Zhang Ping'an said, "Pangu Big Model aims to provide each industry, enterprise, and individual with their own expert assistants, making work more efficient and easy. We always adhere to the strategy of AIforIndustries and continue to move forward on the path of deepening the industry. I firmly believe that Big Model will reshape thousands of industries, and every developer will be a hero who changes the world

Pangu Model 3.0: Reshaping a Thousand Elements and Various Industries

The Huawei Cloud Pangu Big Model has been focused on serving the industry from the beginning. The Pangu Big Model 3.0 released at this conference is also a series of industry oriented big models, including a "5+N+X" three-layer architecture:

The L0 layer includes five basic models: natural language, vision, multimodality, prediction, and scientific computing, providing various skill requirements in industry scenarios. Pangu 3.0 provides customers with a series of basic large models with 10 billion parameters, 38 billion parameters, 71 billion parameters, and 100 billion parameters, matching their diverse industry needs in different scenarios, delays, and response speeds. At the same time, we provide a new set of capabilities, including knowledge Q&A, copywriting, and code generation for NLP models, as well as image generation and understanding for multimodal models. These skills can be directly used by customers and partner enterprises. No matter how large the parameter scale of the large model, Pangu provides a consistent set of capabilities.

The L1 layer is an industry wide model, and Huawei Cloud can provide industry wide models trained using publicly available industry data, including government, finance, manufacturing, mining, meteorology, and other large models; You can also train your own proprietary big model for customers based on their own data in the L0 and L1 layers of the Pangu big model.

The L2 layer provides customers with more detailed scenario models, focusing more on specific industry applications or business scenarios such as government hotlines, branch assistants, lead drug screening, conveyor belt foreign object detection, and typhoon path prediction, providing customers with "out of the box" model services.

The Pangu large model adopts a fully layered decoupling design, which can quickly adapt and meet the changing needs of the industry. Customers can load independent datasets for their large models, upgrade the basic model separately, or upgrade the capability set separately. On the basis of the L0 and L1 big models, Huawei Cloud also provides customers with a big model industry development kit. Through secondary training of customers' own data, customers can have their own exclusive industry big models. At the same time, according to different data security and compliance demands of customers, Pangu Big Model also provides diversified deployment forms of Public cloud, big model cloud zone and Cloud computing#Hybrid cloud.

Pangu was born for the industry, so we must think for the sake of the industry and better serve customers from various industries, "said Zhang Ping'an. Nowadays, the Pangu model has played a huge role in many industries such as finance, finance, manufacturing, pharmaceutical research and development, coal mining, and railways.

In the field of government affairs, Huawei Cloud, together with the Futian District Government Affairs Service Data Management Bureau, launched Xiaofu, a smart assistant of Futian government affairs based on the Pangu government affairs model, which can accurately understand the people's consulting intentions, change the traditional one network model, transform the people's words into the language of government affairs, and make the city more loving. The Pangu Government Big Model has finely tuned over 200000 pieces of government data, including 12345 hotlines, policy documents, government encyclopedias, etc., and has mastered rich industry knowledge such as laws and regulations, business processes, etc. Based on the Pangu government model, Xiaofu embodies its industry, professionalism, leadership, and benefits to the people in the One Network Service.

In the field of coal mining, the Pangu Mining Large Model has been used on a scale of 8 mines nationwide. A large model can cover more than 1000 sub scenarios in the mining, excavation, machinery, transportation, transportation, washing and other business processes of coal mines, allowing more coal miners to work on the ground. This not only makes the working environment more comfortable for coal miners, but also greatly reduces safety accidents.

In the field of railways, the Pangu Railway Large Model can accurately identify 67 types of freight cars and over 430 types of faults running on the current network, with a fault free image screening rate of up to 95%. It has become a powerful digital assistant around freight train inspectors, freeing them from the daily "sea of images" of millions of images.

In the field of meteorology, the Pangu Meteorological Model is the first AI prediction model with accuracy exceeding traditional numerical prediction methods, and the prediction speed has also been significantly improved. The original prediction of the path of a typhoon in the next 10 days requires 5 hours of simulation on a high-performance Computer cluster with 3000 servers. Based on the pre trained Pangu meteorological model, researchers can obtain more accurate prediction results within 10 seconds by configuring a single card on a single server through AI inference.

In the financial field, the Pangu Financial Big Model can pre train various operations, policies, and case documents of banks. It can automatically generate process and operation guidance for counter staff based on customer issues, reducing the original average of 5 operations to 1, and shortening the completion time by more than 5 minutes. The Pangu model provides hundreds of thousands of branch tellers with their own intelligent assistants.

In the manufacturing field, in the past, it used to take more than 3 hours to develop a device allocation plan for a single production line to complete a production plan for one day. After learning various device data, business processes, and rules on the Huawei production line, the Pangu Manufacturing Big Model can accurately understand the business requirements and call the Tiancho AI solver plugin. In one minute, it can make production plans for the next three days.

In the field of Drug development, the research and development of the original new drug takes an average of 10 years and costs 1 billion dollars. Pangu's drug molecular model helped Professor Liu Bing's team in the First Affiliated Hospital of Xi'an Jiaotong University to discover the world's first new target and new category of antibiotics in 40 years, and shortened the lead Drug development cycle to one month, reducing research and development costs by 70%.

Taking root in technology and achieving ultimate efficiency, creating another pole of AI in the world

The innovation of large models is not only the innovation of the model itself, but also relies on the innovation of various root technologies of AI. Huawei has built the AI computing cloud platform based on Kunpeng and Shengteng at the bottom, as well as the Heterogeneous computing architecture CANN, the full scene AI framework MindSpot, AI development production line ModelArts, etc., providing distributed parallel acceleration, operator and compiler optimization, cluster level communication optimization and other key capabilities for the development and operation of large models. Based on Huawei's AI root technology, the training efficiency of large models can be tuned to 1.1 times that of mainstream GPUs in the industry.

Computational power is the foundation for training large models. At this conference, Zhang Ping'an announced that the Ascension AI cloud service with a single cluster of 2000PFlops computing power will be launched simultaneously at Huawei Cloud's Ulanqab and Gui'an AI computing power centers. In addition to supporting Huawei's full scenario AI framework, Shengteng AI cloud service also supports mainstream AI frameworks such as Python and Tensorflow. Meanwhile, 90% of the operators in these frameworks can be smoothly migrated to the Ascension platform through Huawei's end-to-end migration tools. For example, Meitu migrated 70 models to Shengteng in just 30 days, while Huawei Cloud and Meitu team optimized over 30 operators and accelerated the process in parallel, resulting in a 30% improvement in AI performance compared to the original solution.

In addition, GPU failures are often encountered during the training process of large models, and R&D personnel have to frequently restart training, which takes a long time and incurs high costs. The Shengteng AI cloud service can provide longer and more stable AI computing power services, with a 30 day long stability rate of 90% for thousand card training and a breakpoint recovery time of no more than 10 minutes.

In order to help global customers, partners, and developers train and use big models, we are committed to creating another pole of world AI for global customers, providing new choices for all AI developers, "said Zhang Ping'an.

Deep integration, reconstruction of Pangu large model, product innovation

In addition to the practical implementation in Qianxing Baiye, the Huawei Cloud Pangu Big Model has also deeply integrated Huawei Cloud's product services and restructured product innovation.

For example, in data services, using the copywriting and code generation techniques of the Pangu model can improve the efficiency of data writing and front-end code writing, greatly shortening the launch and empowerment cycle of new products. In cloud customer service, by embedding industry knowledge base and intention mining ability in dialogue Q&A, priority AI answering throughout the entire process is achieved, improving customer service work efficiency by 30%. In BI, automatic recommendation from SQL to visual charts is achieved through NL2SQL and AutoGraph intelligent routing. Through multiple rounds of natural language interaction, everyone can easily insight into business details from the data. In cloud search, multimodal embedding and NL2API technology are used to achieve a wide range of scene searches such as videos, texts, and graphs. With powerful semantic understanding and generalization capabilities, the search accuracy is improved by 15%.

At the same time, Huawei Cloud has combined the CodeArts development tool with the Pangu model and officially released CodeArtsSnap, an intelligent programming assistant for developers. The tool has trained 76 billion lines of selected code and 13 million Technical documentation, and has three core functions: intelligent generation, intelligent question answering, and intelligent collaboration. It can realize code generation with one conversation, automatic annotation and test case generation with one click, and intelligent deployment with one command, so that every software developer has his own programming assistant.

In addition, in order to enable enterprises to build stronger content innovation capabilities in the AI era, Huawei Cloud empowers MetaStudio's digital content production line with the Pangu Basic Big Model, creating the Pangu Digital People's Congress Model, providing model generation and model driven services, and has already used 200000 hours of audio and video data for pre training. Based on these two major services, developers can quickly generate and drive digital human models, empower industry applications such as online education, entertainment live streaming, and enterprise conferences, and enable every enterprise employee to achieve "digital human freedom". For example, users only need to upload a 20 second personal video on the service page of Huawei Cloud MetaStudio to quickly generate personalized digital person explanation videos. The work completed by the past three research and development personnel in three days can now be completed in just three minutes.

Open and Fly Together, Collaborate with Customer Partners to Win the Big Model Ecology

Huawei Cloud is committed to building an open and win-win global ecosystem centered on developers. At present, the number of global developers for Huawei Cloud has exceeded 4.6 million, and more than 10000 products have been listed on cloud stores. Huawei Cloud provides an easy-to-use and reliable large model tool kit, a massive multi industry scenario API called Kaitian aPaaS, and a dedicated community for large models that includes rich and high-quality courses and technical certifications. We hope to work with developers and partners to explore the innovative path of combining Pangu large models with the industry.

The development of large models cannot be separated from the continuous input of high-quality data. Huawei Cloud, together with the China Public Relations Association, the Culture Big data Industry Committee, and a number of partners, jointly proposed to establish a big model high-quality data alliance. The Alliance will gather open data from all member units, create high-quality data sets covering thousands of industries, and promote the vigorous development of industry models.

Developers are the core force driving digital innovation. At this conference, the 2023 Huawei Developer Competition with the theme of "Unlimited Creativity" was officially launched. As a top competition in Huawei's ICT field, the Huawei Developer Competition has opened two major tracks: cloud base and industry, covering five major regions in China, Asia Pacific, Europe, North Africa, and Latin America, allowing developers to showcase their skills in the field of artificial intelligence and achieve innovative application development.

Disclaimer: The content of this article is sourced from the internet. The copyright of the text, images, and other materials belongs to the original author. The platform reprints the materials for the purpose of conveying more information. The content of the article is for reference and learning only, and should not be used for commercial purposes. If it infringes on your legitimate rights and interests, please contact us promptly and we will handle it as soon as possible! We respect copyright and are committed to protecting it. Thank you for sharing.(Email:[email protected])