From data to large model application, on November 25th, Hangzhou Yuanchuang Conference was held to share development tips.

Calculation, for incalculable value

2023 Hangzhou Yunqi Conference comes to a shocking conclusion

Since 2015, cloud computing has supported mobile Internet innovation

In the AI era, continue to support the innovation and dreams of all developers

When the theme of the conference returns again

Let us also open the space-time tunnel

Let’s review the 72-hour journey to Yunqi

Create the most open cloud in the AI era

With the rapid development of artificial intelligence (AI) large model technology, the era of intelligence is beginning. AI will become a new productivity in all walks of life and put forward higher requirements for computing power. From underlying computing power to AI platforms to model services, Alibaba has increased investment in research and development to promote comprehensive technological upgrades and innovations in Alibaba Cloud.

"We want to build the most open cloud in the AI era." said Tsai Chongxin, chairman of the board of directors of Alibaba Group.

Cloud computing is an important infrastructure for the digital economy and even the whole society. Currently, 80% of the country's technology companies and more than half of the large AI model companies run on Alibaba Cloud.

"We firmly believe that without openness, there will be no ecology, and without ecology, there will be no future. At the same time, we must always climb to the peak of technology. Only by standing on more advanced and stable technical capabilities can we have greater confidence in opening up." He said .

Through this "most open cloud in the AI era", it becomes easier and cheaper to develop and use AI, helping all walks of life, especially small and medium-sized enterprises, to turn AI into huge productivity .

Facing the intelligent era, Alibaba Cloud will upgrade its cloud computing system through full-stack technological innovation from underlying computing power to AI platform to model services to create the most open cloud in the AI era.

The third wave of cloud computing

The combination of artificial intelligence and cloud computing will bring about the third wave of cloud computing . It will not be completed in one or two years. It may give us ten or decades to allow enough innovation to be implemented in The cloud computing era can be invented.

Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud, delivered a speech on the theme of "The Third Wave of Cloud Computing", reviewing the development of cloud computing. In the first wave, cloud computing changed today's global Internet, including Netflix and MiHoYo. The companies represented are 100% on the cloud from day one.

In the second wave, “traditional enterprises” began to use cloud computing. Wang Jian said, "I put 'traditional enterprises' in quotation marks without any derogatory meaning. It means that the enterprises we have seen in the past have begun to use cloud computing. I personally think it is the beginning of the second wave, because these enterprises were in that era. It seems to be a pioneer in IT technology.”

Among them, the most iconic event is the 2022 Beijing Winter Olympics, where cloud computing carried the core system of the event for the first time. Wang Jian said that the relationship between cloud computing and the Olympic Games is not simply about using technology or not using technology. Its significance is like the first use of alternating current at the World Expo in Chicago.

In the third wave of cloud computing, artificial intelligence and cloud computing will have a concentrated manifestation and explosion in 2023. The emergence of the GPT model has made computers have a very important impact on the revolution of technological innovation.

Wang Jian believes that if we look at the electricity revolution more than a hundred years ago, the relationship between cloud computing and GPT is the relationship between electricity and electric motors . In the future, the computing power of cloud computing will actually be consumed by these motors in the intelligent era, that is, by models.

Regarding generative AI, even a leading company like Nvidia cannot avoid one thing in the end, so when it was released, he said, "In the end, all these calculations will be provided in the cloud . " I think this is also the vitality of an industry. Just like today's electricity may not be the same as the electricity Edison talked about, but as a public service and as an infrastructure, electricity has a very long-lasting vitality. .

Yunqi’s big release

#01 Tongyi Qianwen 2.0: Large-scale model with hundreds of billions of parameters

Alibaba Cloud officially released Tongyi Qianwen 2.0, a large model with hundreds of billions of parameters. In 10 authoritative evaluations, the comprehensive performance of Tongyi Qianwen 2.0 exceeded GPT-3.5 and is accelerating to catch up with GPT-4. In addition, Tongyi Qianwen APP is officially launched in major mobile application markets, and everyone can directly experience the latest model capabilities through the APP.

At the same time, in order to make it easier for large models to be integrated and implemented in various industries, eight major industry model groups based on Tongyi large model training have been launched. They are:

• Tongyi Lingma - Intelligent Coding Assistant

• Tongyi Zhiwen-AI reading assistant

• Tongyi Tingwu - AI assistant for work and study

• Tongyi Stardust-Personalized character creation platform

• Tongyi Dianjin-Intelligent investment research assistant

• Tongyi Xiaomi – Intelligent customer service

• Tongyi Renxin - personal health assistant

• Tongyi Farui-AI Legal Consultant

#02 Artificial Intelligence Platform PAI: New Upgrade

In 2009, Alibaba Cloud proposed the concept of "the data center is a computer". Today, the AI era requires such a technical system even more.

To this end, Alibaba Cloud has newly upgraded its artificial intelligence platform PAI . The bottom layer of PAI adopts the HPN 7.0 new generation AI cluster network architecture, which supports cluster scalability up to 100,000 cards. The acceleration ratio of ultra-large-scale distributed training is as high as 96%, far exceeding the industry level; in large model training tasks, it can save More than 50% of computing power resources, leading the world in performance.

In addition to Tongyi Large Model, half of China's major model companies run on Alibaba Cloud. A large number of leading companies and institutions such as Baichuan Intelligence, Zhipu AI, Zero One Wish, Kunlun Wanwei, vivo, and Fudan University are all trained on Alibaba Cloud. Large model.

#03 One-stop large model application development platform: Alibaba Cloud Bailian

"Bailian" integrates mainstream high-quality large models at home and abroad, provides model selection, fine-tuning training, security suites, model deployment and other services as well as full-link application development tools, simplifying the deployment of underlying computing power, model pre-training, and Complex work such as tool development.

Developers can develop a large model application in 5 minutes, and "refine" an enterprise-specific model in a few hours. Developers can focus more on application innovation.

Early users such as Langxin Technology, CCTV Network, and AsiaInfo Technology have successfully developed their own exclusive models and applications on the Alibaba Cloud Bailian platform, bringing important innovations to their respective fields. Among them, Langxin Technology is a leading company in the field of energy technology and has long been committed to services and development in the field of electric energy consumption.

#04 Towards an autonomous driving cloud

Stability, security, performance, and cost are still the most important cornerstones of the cloud. After a large amount of investment in R&D, the performance of Alibaba Cloud's basic cloud products has been further improved, the overall availability has moved from 4 9s to 5 9s, and more than 30 cloud products have been intelligently upgraded. At the same time, Alibaba Cloud's serverless process has entered the fast lane. More than 40 cloud products provide serverless capabilities, and the world's first container computing service ACS has been launched.

More than 40 core products are serverless:

As serverless technology continues to evolve, the way developers use the cloud has also changed from centralized R&D and distributed R&D in the past to process R&D on the cloud. Whether there are 20,000 users or 20 million users, the IT architecture built on Serverless can be adaptively scaled, automatically expanding in seconds during peak peaks and automatically shrinking during peak and valley periods.

Alibaba Cloud's serverless evolution has also entered the fast lane. As of October 2023, it has more than 40 serverless cloud products, including function computing FC, serverless application engine SAE, PolarDB database, open source big data platform E-MapReduce, etc.

Among them, Function Compute has been upgraded to version 3.0, cold start has been reduced to seconds, resource costs have been reduced by 92%, complexity has been reduced by 40%, and Tongyi product series, Magic Community, HuggingFace, etc. are deeply integrated, and AI applications can be deployed with one click and the deployment time is Reduced by more than 80%.

At the meeting, Alibaba Cloud also announced price cuts for its entire range of Serverless products. The Serverless versions of big data, databases, middleware, etc. have all seen price cuts ranging from 30% to 70%, making them more attractive.

The world’s first container computing service:

Container technology is one of the hottest technologies in the IT industry today, with 64% of end users using K8s and related container technologies in production. As containerization and cloud migration become the new normal, how to keep up with business needs and avoid resource waste has become a new challenge.

As the earliest cloud vendor in China to lay out container technology, Alibaba Cloud took the lead in launching the world's first container computing service (Alibaba Cloud Container Compute Service, hereinafter referred to as ACS), which delivers computing power in a serverless form. Users do not need to pay attention to the underlying nodes and clusters. Configure to create pod resources and deploy applications.

ACS adopts Alibaba Cloud's self-developed scheduling technology, pioneering computing power allocation and resource reuse, supporting computing power switching between user loads and cloud products, improving resource utilization, and saving enterprises 20% of resource costs. ACS also supports flexible usage modes such as a maximum 1:8 resource ratio and on-demand elasticity and resource reservation to ensure daily needs and achieve the perfect combination of flexibility and certainty.

#05 The first SSD main control chip, Zhenyue 510

Pingtou Ge released its first SSD main control chip Zhenyue 510. This chip is deeply customized for cloud computing scenarios and achieves 4μs ultra-low latency, which is more than 30% lower than the mainstream in the industry. The bit error rate is as low as 10^-18, which is lower than the industry standard. The benchmark is an order of magnitude ahead.

Zhenyue 510 will be the first to be deployed in Alibaba Cloud data centers and can be used in business scenarios such as AI, online transactions, big data analysis, high-performance databases, and software-defined storage.

"Computing power, network power, and storage power jointly build the foundation of cloud computing. Computing power and network have evolved rapidly in the past few years. Storage technology represented by SSD must also be designed to be integrated with cloud computing software systems to provide enterprise users with Bringing a better storage experience." said Pingtouge Product Director.

"AI More Possibilities" Audio and Video Technology Exhibition

At this year's Yunqi Technology Exhibition, Alibaba Cloud Video Cloud centered around the "Digital Human x Intelligent Editing and Synthesis" and "AIGC Media Content Production" experience booths, fully presenting the rich scenarios of audio and video technology innovation and integration in the AIGC era.

" Digital Human _

"AIGC Media Content Production" mainly shows that the video cloud seamlessly integrates large models with audio and video technology frame granularity, and builds an intelligent media service capability set around the content production field in live video and video on demand scenarios.

It includes a virtual studio that uses multi-channel real-time keying, virtual background generation and real-time dynamic rendering technology . Experiencers and virtual digital people can interact in real time on the same stage, creating an imaginary space station where virtuality and reality coexist.