Shenzhen Business Daily Reading Client Reporter Tu Jingyu
On April 23, SenseTime held a technical exchange day and released the industry's first "cloud, device, edge" full-stack large model product matrix to meet the application needs of different scale scenarios, and newly upgraded the "SenseNova 5.0" large model system, whose comprehensive capabilities are fully benchmarked against GPT-4 Turbo, and the technology leads the comprehensive leap from generative AI to industrial landing, so as to realize the on-demand use of large models.
Under the principle of Scaling Law, the most basic law of AI development, SenseTime continues to seek the best data ratio and establish a data quality evaluation system to promote its own large model research and development, while also providing industry partners with large model training, fine-tuning, deployment and various generative AI capabilities and services.
Xu Li, Chairman and CEO of SenseTime, said, "Under the guidance of the law of scale, SenseTime will continue to explore the KRE three-layer architecture (knowledge-reasoning-execution) of large model capabilities, and constantly break through the boundaries of large model capabilities. ”
The new multi-modal interaction has greatly improved the ability of dual study in arts and sciences
Since its first release in April last year, SenseTime's "RiRixin SenseNova" large model system has officially launched five major version iterations. Based on more than 10TB of tokens trained and covering a large amount of synthetic data, the new "RiRixin SenseNova 5.0" (hereinafter referred to as RiRixin 5.0) adopts a hybrid expert architecture, and the context window can be effective to about 200K during inference.
This update mainly focuses on enhancing knowledge, mathematics, reasoning and code capabilities, comprehensively benchmarking GPT-4 Turbo, and reaching or surpassing GPT-4 Turbo in mainstream objective evaluations.
In terms of liberal arts ability, the creative writing ability, reasoning ability and summary ability of "Ririxin 5.0" have been greatly improved, and after the same Chinese knowledge is injected, you can get better understanding and summary and Q&A, providing strong assistance for vertical application scenarios such as education and content industry.
"RiRixin 5.0" and GPT-4 answer interesting reasoning questions: "Mom made Yuanyuan a cup of coffee, and after Yuanyuan drank half a cup, she filled it with water, and then she drank another half cup, and then filled it with water, and finally drank it all." Ask Yuanyuan whether he drinks more coffee or more water?" and "RiRixin 5.0" answered correctly.
In terms of science capabilities, the mathematical and physical capabilities, code capabilities and reasoning capabilities of "Ririxin 5.0" have reached the industry-leading level, providing a solid foundation for the implementation of scenarios such as finance and data analysis.
Another core indicator of the "RiRixin 5.0" is multimodal capability, and the graphic and text perception ability of SenseTime's multi-modal large model has reached the world's leading level, ranking first in the comprehensive score of MMBench, the authoritative comprehensive benchmark test of multi-modal large model, and leading results in many well-known multi-modal lists MathVista, AI2D, ChartQA, TextVQA, DocVQA, and MMMU.
"RiRixin SenseNova 5.0" also achieves more excellent multi-modal capabilities at the application product level, supports the analysis and understanding of high-definition long graphs, and the interactive generation of Wensheng diagrams, as well as complex cross-document knowledge extraction and summary Q&A display, and also has rich multi-modal interaction capabilities.
The device-side model ranks first in the industry, and the enterprise-level application all-in-one machine is launched on the side
With a forward-looking insight into the future trend of the expansion of centralized computing power demand to the device side and the AI demand of enterprises at the edge, SenseTime has launched the industry's first full-stack large-scale model product matrix of "cloud, device, and edge", including the "SenseTime device-side large model" applied to terminal equipment, and the edge product "SenseTime enterprise-level large model all-in-one" for multiple fields such as finance, code, healthcare, and government affairs.
This year is the first year of the application of device-side large models, and in order to meet the application needs of mobile terminal users for large model technology, SenseTime has launched the Ririxin End-side Large Model, which achieves the best performance at the same scale and leads the cross-level scale in an all-round way.
SenseTime has also launched a device-cloud collaboration solution, which can give full play to the respective advantages of devices and clouds through intelligent judgment collaboration, and offload to the cloud for processing when it is necessary to search or process complex scenarios online, with device-side processing accounting for more than 80% of some scenarios, thereby significantly reducing inference costs.
The inference speed of SenseTime's device-side large language model has reached the fastest in the industry, achieving an average generation speed of 18.3 words/s on the mid-range platform and 78.3 words/s on the flagship platform.
The diffusion model can also achieve the fastest inference speed in the industry on the device side, and the inference speed of the end-side LDM-AI expansion technology is less than 1.5 seconds on a mainstream platform, which is 10 times faster than that of the cloud app of a competitor, and supports the output of high-definition images of 12 million pixels and above, and supports fast image editing functions such as equal ratio expansion on the device, free expansion and rotation expansion on the device.
In response to the growing demand for AI applications at the edge of key industries such as finance, code, healthcare, and government affairs, SenseTime officially launched an enterprise-level large model all-in-one machine, which can support both enterprise-level 100-billion-yuan model acceleration and knowledge retrieval hardware acceleration, realize localized deployment, and use it out-of-the-box, lowering the threshold for enterprise application large models. Compared with similar products in the industry, the inference cost is reduced by 80%, the retrieval is greatly accelerated, and the CPU workload is 50%.
Work with ecosystem partners to innovate product applications in the AI 2.0 era
At the event, SenseTime also invited a number of guests from ecological partners such as Kingsoft Office, Haitong Securities, Xiaomi, China Literature Group, and Huawei to discuss and exchange the application and prospects of large model technology in different fields such as office, finance, and travel.
Since 2023, SenseTime has reached an in-depth cooperation with Kingsoft Office, based on the excellent code generation and tool invocation capabilities of the "RiRixin" large model, to help WPS 365 build a new office productivity platform that releases scene capabilities more efficiently, and build an exclusive "enterprise brain" for enterprises. Zhang Qingyuan, CEO of Kingsoft Office, said: "In the office application scenario, the performance of the SenseTime model is very good, which can help our users solve complex problems in the office and improve efficiency. ”
In the financial sector, Haitong Securities and SenseTime jointly released a multi-modal full-stack model of the financial industry, in which the two parties promoted business implementation in the fields of intelligent customer service, compliance and risk control, code assistance, and business office assistants, and jointly researched cutting-edge scenarios in the industry such as robo-advisors and public opinion monitoring, so as to open up the full-stack capabilities of the large-scale model landing in the securities industry. Mao Yuxing, Deputy General Manager and Chief Information Officer of Haitong Securities, said: "Through our cooperation with SenseTime, we have realized the digital and intelligent transformation of Haitong Securities by using large-scale model technology, and in the future, we will combine full-stack AI capabilities to carry out business processes, interactive transformation and digital intelligent business system reconstruction. ”
In the personal travel scenario, SenseTime's large model technology is applied to the smart cabin of the Xiaomi Auto SU7, which has recently become popular in the market. Wang Gang, general manager of Xiaomi Xiaoai, said in a dialogue with Wang Xiaogang, co-founder and chief scientist of SenseTime, "SenseTime's cloud-edge-end full-stack combination can well empower and adapt to Xiaomi's IoT ecosystem. We hope to work with SenseTime to create a more intelligent product experience for our users. ”
In addition, today SenseTime also released a large industry model based on Ascend to jointly build an industry ecosystem for finance, healthcare, government affairs, code, and other large models.
In terms of its own application, SenseTime's "Ririxin SenseNova 5.0" has been updated in Miaohua, Ronin, Gewu, Qiongyu, Dayi, Little Raccoon Family and other products.