With the rapid development of emerging technologies such as cloud native, AIGC and large models, we are ushering in the beginning of the intelligent era. Cloud computing is also entering the 3.0 era, which is the stage of deep integration of cloud computing and artificial intelligence. At this stage, artificial intelligence technology has become the core driving force to further unleash the potential of cloud computing.
On January 4, Kingsoft Cloud held a media communication meeting on "Cloud Artificial Intelligence·New Opportunities in the Era". Qian Yifeng, Vice President of Kingsoft Cloud, Sun Xiao, Head of Kingsoft Cloud Public Cloud Product Center, Kingsoft Cloud Artificial Intelligence and Big Data Xu Yinfei, head of the product center, shared Kingsoft Cloud’s technical history and implementation practices in terms of basic capabilities, platform capabilities and model services at the meeting. Through self-research and upgrading of technology, Kingsoft Cloud has initially built a full-stack cloud computing system for the civilian field in the era of artificial intelligence.
In order to improve artificial intelligence service capabilities, we need to lay a solid foundation and conduct comprehensive upgrades.
The global digitalization wave is surging, and cloud computing, as an important supporting foundation for the digital economy, is playing an increasingly important role. Among them, intelligence, as an important concept in the cloud computing 3.0 era, has become an important driving force for the upgrading of digital infrastructure. A report released by IDC, a well-known research organization, in 2023 shows that in the next five years, the next generation of artificial intelligence driven by large models and generative artificial intelligence is expected to drive the overall cloud industry through the downward stage and return to the era of high growth.
Following the trend, Kingsoft Cloud deeply explores cutting-edge technologies, upgrades core computing, storage, network and other products around AIGC, and fully embraces artificial intelligence from multiple dimensions. In terms of computing, Kingsoft Cloud’s seventh-generation cloud server X7 is equipped with the fourth-generation Intel® The performance of the previous generation is improved by 50%. Integrating Kingsoft Cloud's independently innovative acceleration technology, cloud server X7 can effectively improve model inference performance.
According to the different heat levels of data from extremely hot to extremely cold, Kingsoft Cloud Object Storage has covered storage products with different access hotness such as standard storage, low-frequency storage, deep low-frequency storage, archive storage, deep cold archive storage and all-flash storage. Among them, Kingsoft Cloud Object Storage KS3 Extreme Speed can provide up to 1Tbps/PB of redemption bandwidth, which is hundreds of times better than object storage based on mechanical hard disks (HDD). It can provide AIGC, storage and computing separation, high-performance computing and other scenarios. Powerful storage solution.
For large model application scenarios, Kingsoft Cloud launches Mutual Trust Virtual Private Network (referred to as "Mutual Trust VPC"). Compared with standard VPC, mutual trust VPC has stricter control over communication behavior in and out of the VPC, helping customers solve compliance and mutual trust issues. For application layer load scenarios such as HTTP and HTTPS, the company launched application load balancing ALB (Application Load Balancer), with a single instance supporting up to 1 million QPS. Integrated with cloud-native scenarios, when customers use Kingsoft Cloud Container Service KCE products or build their own K8S clusters, ALB can be deployed as an Ingress to provide network traffic scheduling services for the business.
Currently, Kingsoft Cloud has established a complete technical system. This system takes cloud product infrastructure as its basic capabilities and cloud product platform services as its core technical capabilities. It provides assistance to artificial intelligence customers through the MaaS mutual trust reasoning zone solution and Kingsoft Cloud Qingzhou Zhiwen.
In order to help players pass the levels faster, let’s talk about some acceleration techniques that can help you become a powerful puzzle expert.
2023 is called the "first year" of large models. Public data shows that from the beginning of the year to the end of the year, the number of domestic large models exceeded 200. Contrary to the exponential growth in the number of "large models", there are relatively few large models that are actually called. In this wave of large models, Kingsoft Cloud adheres to its neutral positioning, gives full play to its base and platform capabilities, and becomes a facilitator of large models.
After gaining insight into the pain points of model supply and demand sides, Kingsoft Cloud took the lead in launching the MaaS Mutual Trust Reasoning Zone solution (hereinafter referred to as "MaaS 1.0") in June last year to establish mutual trust between large model manufacturers, users and Kingsoft Cloud. Solve the problem of mutual trust between models and data. At this communication meeting, Kingsoft Cloud released the MaaS Mutual Trust Reasoning Zone Solution 2.0 (hereinafter referred to as "MaaS 2.0").
Based on MaaS 1.0, MaaS Mutual Trust Reasoning Zone Solution 2.0 uses Kingsoft Cloud IaaS and PaaS as the base, which can realize one-click deployment of LangChain on the cloud, and connects to multiple large ecological cooperation business models and open source large models by default. It also supports Embedding models including BGE, Bert, etc., can seamlessly connect to Kingsoft Cloud's fully managed vector database Milvus, and provides an easy-to-use, safe and trustworthy one-stop inference application deployment platform for enterprise developers. In addition, MaaS 2.0 supports RAG large model scenario applications including model reasoning and knowledge base construction through standardized API interfaces and Web front-end interfaces. In order to further enhance the security of running on the cloud, MaaS 2.0 also provides a container service encrypted image solution. Relying on Kingsoft Cloud bare metal service, it realizes the full process model of image encryption, encrypted image upload, and decrypted image operation in Kingsoft Cloud container service. safe operation.
At the same time, in order to meet the needs of industry customers, Kingsoft Cloud explores new opportunities for enterprise empowerment in the era of large models, and formulates a “one, three, one, four” product panoramic plan around enterprise-level knowledge assistant scenarios, that is, a set of capabilities (Kingsoft Cloud Qingzhou Zhiwen), three models (industry language model, text fragmentation and Embedding model), one platform (Kingsoft Cloud Hanhai Platform) and four major functions (fine-tuned reasoning, data acceleration, intelligent retrieval and document intelligence), around generative artificial intelligence Intelligent construction of full-stack capabilities for application implementation, model training fine-tuning, and platform support. Based on the “1314” plan, Kingsoft Cloud will export a number of atomic capabilities to application customers and platform customers respectively. At present, Kingsoft Cloud Qingzhou Zhiwen Knowledge Assistant product has completed the application and construction of core technologies such as Embedding model, multi-channel recall algorithm and intelligent data slicing model, and plans to prioritize its implementation in public service and legal scenarios.
The combination ofand artificial intelligence gives the cloud more room for imagination and brings new productivity to various industries. With the further improvement and implementation of technology, cloud computing will usher in stronger and longer vitality. Kingsoft Cloud will continue to “train its internal skills” around customer needs and work with ecological partners to lay out the future with differentiated strategies.
The above is the detailed content of Kingsoft Cloud upgrades its full-stack cloud computing system to become a 'facilitator” on the cloud in the era of artificial intelligence. For more information, please follow other related articles on the PHP Chinese website!