SenseTime released the "DayDay New Sensenova 4."0", multi-dimensional comprehensive upgrade of the large model system.
Every day the new sensenova 40 "has more complete knowledge coverage, more reliable reasoning ability, superior long-term understanding, more stable numerical reasoning ability and stronger success ability, and cross-modal interaction. The general version (SenseChat V4) supports a 128K contextual window length, and the overall evaluation score level is comparable to GPT 4, compared with GPT 35 has achieved total transcendence. (API application**: SenseTime has also taken the lead in launching the new Function Call & Assistants API version of the large language model that supports different modal tool calls, providing a convenient bridge for the connection between advanced large models and various application service tools, and significantly reducing the threshold for developers to use large models.)
SenseTime has also transformed its advanced large model capabilities into product applications in actual scenarios, and completed the development and released data analysis tools based on the newly released Function Call & Assistants API for the new large language model"Office Little Raccoon".
At present, with its abundant AI technology capabilities, SenseTime's "Sense Nova" large model system has opened up various AI applications with low cost and high efficiency, providing the following services: SenseTime New Consultation Language Model for Office Scenarios - Data Analysis Version, Everyday New Consultation Language Large Model for Medical Scenarios - Medical Version "Big Doctor", Ririxin Consultation Multimodal Large Model for Autonomous Driving and Industrial Scenarios, and Everyday New for Creative Scenes· Rich tools such as large models of Wensheng diagrams in seconds.
SenseTime's "New Consultation Large Language Model" has established in-depth cooperation with more than 500 customers in many vertical industries such as finance, mobile phones, medical care, automobiles, real estate, energy, media, and industrial manufacturing, accelerating the intelligent transformation of industry enterprises.
Next, SenseTime's "RiRixin Sensenova" large model system and related products and tools provide a precise focus for the realization of general artificial intelligence (AGI), help all scenarios and multiple industries to achieve "large model +", and broaden the application boundaries of large models.
"Big new before the holiday"! Ultra-rich large model system, AI capabilities can be used at any time
Every day the new sensenova 4"0" provides a variety of flexible API interfaces and services, and developers can easily call the AI technical capabilities of the Sensenova large model according to the actual application needs, so as to realize various AI applications at lower cost and higher efficiency.
The basis for improving the performance of large models lies in the enhancement of language modeling capabilities. Every day the new sensenova 40" brand new upgradeEveryday New Consultation Large Language Model - Universal Version (sensechat v4).At present, it supports 4K, 32K, and 128K tokens windows, expanding the application scope of the model. SenseChat V4 significantly improves the general ability in the fields of knowledge comprehension, reading comprehension, comprehensive reasoning, mathematics and science, and long text comprehension, and the overall assessment score level is comparable to GPT-4, and the performance of reasoning and programming is even better than GPT-4. In the test of the authoritative test set Humaneval Coding, the first-time pass rate reached 756% (GPT-4 is 74.)4%）。
Based on the full test of the large model evaluation platform OpenCompass, the large language model SenseChat V4 is comparable to GPT-4 in terms of overall evaluation scores
SenseTime NewSenseChat-DataAnalysis v4Published in the data test set (1000+ questions) in the data analysis scenarioTake 8571% correct rate over GPT-4It supports the understanding of complex, multi-file and multi-file, realizes common data analysis tasks such as data cleaning, data operation, comparative analysis, trend analysis, analysis and visualization, and empowers multiple application scenarios such as financial analysis, business analysis, sales, market analysis, and macro analysis.
SenseTime launched a new dayConsultation Language Model - Medical Version "SenseChat-Medical V4".In this update, it achieves stronger multi-round dialogue and context understanding capabilities, as well as richer tool invocation capabilities, effectively realizes professional medical Q&A and complex medical task reasoning, and supports intelligent interpretation and interactive Q&A of more modal medical documents. "Big Doctor" has achieved the second comprehensive score in two industry authoritative evaluations - the 2023 professional pharmacist examination large model evaluation and the Chinese medical large language model open evaluation platform Medbench, and its performance is close to GPT-4, of which the performance of two subdivisions in the evaluation of the professional pharmacist examination large model has surpassed GPT-4.
In the results of the 2023 professional pharmacist examination large model evaluation, the comprehensive score of "big doctor" ranked second, and the performance surpassed GPT-4 in two subdivisions
The web-end display page of Dayi covering 13 medical scenarios.
Multimodality is an important technological evolution direction for large AI modelsDiscuss the multi-modal large model (sensechat-vision v4).Newly launched, it has 30 billion parameters, and its understanding ability is at the world's leading levelRanked first in the overall score on the MME Benchmark, an authoritative benchmark setThe overall score is 21995 (4 over GPT-1926.)57), which has been applied in many practical scenarios such as intelligent driving, intelligent cabin, and power industry, helping the industrial upgrading of thousands of industries.
The MME Benchmark can systematically comprehensively evaluate the model's ability from 14 dimensions, including positioning, celebrity recognition, scenic spot recognition, OCR, and mathematical calculation.
The multimodal model can answer complex tasks such as content understanding in traffic scenarios.
SenseTime's self-developedSensemirage V4Compared with the previous version, the number of parameters has been increased to tens of billions, and the semantic comprehension ability and image texture detail performance have been significantly enhanced through the optimization of algorithms such as Mixture of Text Experts and Spatial-Aware CFGIt can reach the level of cinematic poster generation。At the same time, combined with the Adversarial Distillation algorithm, the Sensemirage-Turbo V4 version of Miaohua has also been released, which can achieve 10 times the inference acceleration effect compared with the basic version of V4.
Sensemirage V4 is a new upgrade, and the movie-level ** is generated with one click.
The first version of the Function Call & Assistants API to support different modal tool calls is here, an exclusive development assistant in the era of large models
In order to help more developers and related industries make good use of large models more conveniently and efficiently, SenseTime has launched the cutting-edge and abundant capabilities of large modelsNew • Discuss the API version of the large language model Function Call & Assistants
The new Function Call & Assistants API version of the large language model provides developers with a flexible and highly customizable tool calling framework, providing a variety of tools such as online search, explanation, Q&A, graphs, etc., and holding SenseTime's new model system to empower multiple business scenarios.
Compared with other similar products, the new large language model Function Call & Assistants API versionThe first to support tool invocation in different modalitiesIt can support the combined multi-modal interaction and the visual presentation of data analysis execution results, so as to solve more complex problems and make the integration of AI functions in various applications simpler and more effective.
Through the Function Call & Assistants API of the new large language model, you can easily call the built-in tools to achieve multi-modal interaction (note: the display effect has accelerated processing).
The application of innovative scenarios has been accelerated, and SenseTime's "daily new" large-scale model system continues to empower multi-industry upgrading
The emergence of large models has brought about a huge transformation of the human-computer interaction paradigm, relying on the rapid iteration of SenseTime's "daily new sensenova" capability, a new member of the little raccoon family - data analysis tools"Office Little Raccoon"., reshaping the form of application products in the era of large models.
Office Raccoon is a data analysis tool that can be used without programming or complex operations. Through natural language input, Office Little Raccoon combines the intent recognition, logical understanding, and generation capabilities of SenseTime's large model system to automatically transform data into meaningful analysis results and visual charts. Thanks to the strong Chinese comprehension ability of SenseTime's large model system, the product is more suitable for China's data analysis needs.
Office Raccoon Demo Page: Enter ** to easily get the analysis report. Experience Entrance:
Since the launch of SenseTime's "Ririxin Sensenova" on April 10, 2023, there have been more than 3,000 enterprise users, covering industries including the Internet, games, cultural tourism, education, medical and health, finance, and programming.
In the future, SenseTime will continue to promote the construction of the "Sensenova" large model system, realize innovative scenario applications, and continuously promote the front-loading of large models, so that more intelligence can go to more scenarios and industries, and join hands with the industry ecology to move towards the agi era.