Launched the Wenxin Big Model 4.0 API Call Service Test Application for enterprise customers, serving more than 17,000 customers and exploring the application of large models in nearly 500 scenarios in various industries... Since its launch in March this year, Baidu Intelligent Cloud Qianfan Big Model Platform, as the world's first one-stop enterprise-level large model platform, has delivered the only "full score report card" for the industry.
On October 17, Baidu World 2023 was held in Beijing Shougang Park. Baidu Intelligent Cloud announced during the conference that it would comprehensively upgrade its "cloud-intelligence integration" strategy and provide full-stack service solutions for the five types of customer requirements for landing large models; Aiming at AI native application development, the "Qianfan AI Native Application Development Workbench" was released to accelerate the landing of enterprise AI native applications; Released the first AI native app store in China and the first large-model full-link ecosystem support system in China, enabling partners to grow business and jointly build and share a prosperous large-model industry ecosystem.
(Shen Jing, Executive Vice President of Baidu Group and President of Baidu Intelligent Cloud Business Group)
Meet the five major needs of enterprise large model implementation: Baidu Intelligent Cloud upgrades the "cloud-intelligence integration" strategy
Since 2020, Baidu Intelligent Cloud has taken the lead in the industry to propose the "cloud-intelligence integration" strategy, and has carried out many iterative upgrades in terms of technical architecture and services. At this Baidu World Conference, the strategic connotation of "cloud-intelligence integration" has been comprehensively upgraded to "cloud-intelligence integration, deep industry, ecological prosperity, and AI inclusiveness".
Shen Zhen, executive vice president of Baidu Group and president of Baidu Intelligent Cloud Business Group, said that the deep combination of artificial intelligence and cloud computing is the key for enterprises to quickly implement AI native applications, which is also the concept of "cloud-intelligence integration" that Baidu Intelligent Cloud has always advocated and practiced. At present, all applications and services of Baidu Group are running on Baidu Intelligent Cloud based on the "Cloud-Intelligence Integration" technology architecture. In addition, facing the five types of customer needs for landing large models, Baidu Intelligent Cloud has given the best service solutions for the "big model super factory" built based on the Qianfan big model platform.
(Cloud-intelligence integration in the era of large models)
First of all, for customers who only need computing power, Qianfan platform can provide extremely efficient and cost-effective heterogeneous computing power services. In the large model training link that customers are most concerned about, through the distributed parallel training strategy and microsecond-level interconnection capability, the Qianfan platform can achieve an acceleration ratio of 95% for Wanka large-scale cluster training. Through precautions and timely detection, positioning, and solution in the event, the invalid operation of the cluster due to faults and other reasons is avoided to the greatest extent, and the proportion of effective training time is increased, and the effective training time of the Vanka cluster accounts for 96%, fully releasing the effective computing power of the cluster and greatly reducing the customer's computing power and time cost.
In addition, Qianfan platform is also compatible with Kunlun Chip, Ascend, Haiguang DCU, NVIDIA, Intel and other mainstream AI chips at home and abroad, supporting customers to complete computing power adaptation with minimal switching cost.
Based on the AI computing power cluster provided by Baidu Intelligent Cloud, Facewall Intelligence has trained the "Zhihaitu AI" large model and the multimodal large model Luca, accounting for 99% of the effective training time on the kcal cluster, which can achieve efficient convergence of the model training effect while ensuring the continuity of model training. In addition, enterprises such as Zhihu, Hao Future, and Horizon are also using the AI computing power services provided by Baidu Intelligent Cloud to achieve large-scale cluster training and management in a more stable, efficient and economical way.
Secondly, at the model level, for customers who want to directly call existing large models, Qianfan platform manages 42 mainstream large models at home and abroad, and enterprise customers can quickly call the APIs of various large models, including Wenxin large models, to obtain large model capabilities. For third-party large models, Qianfan platform also carries out targeted optimizations such as Chinese enhancement, performance enhancement, and context enhancement. At present, Qianfan platform has served more than 17,000 customers, and the number of large-model API calls continues to increase at a high speed.
Third, for customers who want to carry out secondary development based on existing large models, Qianfan platform provides a full life cycle tool chain and the industry's largest number of 41 high-quality industry datasets for the retraining, fine-tuning, evaluation and deployment of large models, helping customers quickly optimize model effects for their own business scenarios. At present, many leading customers in the industry, including Postal Savings Bank of China, Du Xiaoman, Kingsoft Office, and Hebei High-speed Group, are developing exclusive large models that meet business needs through the toolchain services provided by Qianfan platform.
Fourth, at the application level, some enterprises need to develop AI native applications based on large model services, and a series of capability components and frameworks provided on the Qianfan platform can help enterprises quickly complete application development and flexibly respond to user and market needs.
Finally, some customers want to directly and conveniently purchase mature AI native application products to empower business development.
Empowering Enterprises to Develop AI Native Applications Efficiently: Released the "Qianfan AI Native Application Development Workbench"
In order to meet the needs of enterprises for agile and efficient AI native application development and lower the threshold of AI native application development, Baidu Intelligent Cloud released the "Qianfan AI Native Application Development Workbench", which precipitates the common patterns, tools and processes for developing large-model applications into a workbench, helping developers focus on their own business without having to involve unnecessary energy in the development process. Specifically, Qianfan AI native application development workbench is mainly composed of two layers of services: application components and application frameworks.
Application component services are composed of two types of components, AI and basic cloud, which are componentized encapsulation of the underlying service capabilities, allowing each component to complete a specific function. The "AI component" includes large language model components such as question answering and chain of thought (CoT, Chain of Thought), as well as multimodal components such as text diagrams and speech recognition, while the "basic cloud component" contains traditional cloud service capabilities such as vector databases and object storage.
The application framework selectively connects and combines these components so that they can complete the tasks of a specific scenario relatively completely. At present, the retrieval enhancement generation (RAG) and agent (Agent) provided on the Qianfan platform are commonly used AI native application frameworks. Under each framework, Baidu Intelligent Cloud also provides a wealth of model rooms to support developers to develop AI native applications quickly and efficiently.
(Baidu Intelligent Cloud releases AI native application development workbench)
The Retrieval Augmented Generation (RAG) framework can more efficiently use the knowledge in the company's proprietary domain and make accurate answers according to relevant questions with the help of large models, which is a necessary core capability for AI native applications of professional knowledge question answering.
The conference site also carried out a practical demonstration of how to quickly develop knowledge Q&A applications for Sany Heavy Industry based on the RAG framework: just select the preset RAG framework in the Qianfan AI native application workbench and configure the corresponding parameters to quickly realize the development and launch of the intelligent customer service application on the official website of Sany Heavy Industry. Shen said that building such a "small assistant", even if it needs to process thousands of thousands of words long documents, the cost is only a few hundred yuan; After that, each consultation of the user only costs a few cents.
As a popular application framework in the industry, agents can automatically disassemble tasks given by humans, automatically plan and call various components to complete tasks collaboratively, and at the same time get feedback according to the task completion effect to improve their own capabilities. At present, the Agent framework has been widely used in industry, transportation and other fields.
Based on the Agent framework provided by Qianfan AI native application development workbench, Zhongtian Steel has built an intelligent "enterprise scheduling hub" to realize the automatic perception, decomposition and execution of task instructions. For example, when it is found that the steel output is not up to standard, the user only needs to ask a question once, and the large model can automatically call various resources and APIs managed by the platform, complete BI data retrieval, third-party root cause analysis, etc., find the reason for the non-compliance, adjust the scheduling plan in time and send an email to notify the dispatcher.
(Zhongtian Steel builds an enterprise scheduling hub based on the agent application framework)
Prosperous industrial ecology: China's first AI native app store and full-link large-model ecosystem came out collectively
In order to meet the needs of enterprise customers to conveniently purchase high-quality AI native applications and empower partners' business success, Baidu Intelligent Cloud has launched Baidu Intelligent Cloud Qianfan AI Native App Store, the first AI native app store for one-stop transactions for enterprise customers in China, aiming to establish a convenient and reliable connection channel for the supply and demand sides of AI native applications in the era of large models, greatly improve customer application selection and procurement efficiency, and create more business opportunities for partner merchants. At present, enterprise users can enter the Qianfan AI native app store through the official website of Baidu Intelligent Cloud and purchase their favorite AI native applications.
(Baidu Intelligent Cloud Qianfan AI Native App Store)
At this World Congress, Baidu released the first generative business intelligence product in China - Baidu GBI (Generative Business Intelligence). BI (Business Intelligence) is one of the typical application scenarios that show the "ceiling" of large model capabilities. As an AI native application engine, Baidu GBI has realized a comprehensive reconstruction of the "intelligent questioning" capability based on the Wenxin big model: for any database, query and analysis can be completed in one step through dialogue, which greatly improves the efficiency of data analysis and reduces the requirements for professional data analysis capabilities compared with traditional BI.
Taking enterprise operation data analysis as an example, Baidu GBI can shorten the working time of data analysis and report writing that business data analysts can complete in a dozen days from days to minutes. As a type of AI native application framework product, Baidu GBI can also realize simplified cross-industry and cross-domain capability migration, and quickly implement scenarios such as urban governance analysis, energy consumption insight, industrial equipment operation management, financial investment decision-making, traffic congestion governance, etc., and efficiently realize the development and construction of government and enterprise BI AI native applications.
Kingdee, as one of the first partners to enter the Qianfan AI native app store, and Baidu Intelligent Cloud jointly released Kingdee Cloud Sky GPT on August 8, 2023, and took the lead in launching the industry's first financial big model, helping customers conduct financial management and analysis more efficiently, improve the accuracy and standardization of financial processing, and provide a strong guarantee for the long-term stable development of enterprises.
Big model technology is not only reshaping the service model and product form of cloud computing, but also reshaping the ecological pattern of cloud computing industry, putting forward new requirements for the capabilities and system construction of cloud computing vendors and their partners covering thousands of industries.
In order to help partners achieve business success in the era of big models, Baidu Intelligent Cloud has built and launched the first full-link ecosystem support system for large models in China, and provides 10 comprehensive partners, 100 application partners, more than 10,000 start-ups and agency partners, including Qianfan community, AI native application incubation, sales opportunities, marketing, enabling training, and Qianfan AI native app store.
Yuan Foyu, Vice President of Baidu Group, said, "In this era of rapid change, Baidu must not only brave the no-man's land on the road of large-model technology and commercialization exploration, but also work with many partners to create a prosperous industrial ecology, jointly explore new business models and business opportunities, and achieve true win-win cooperation." ”