Background information
Industrial-level high-performance AI models faces challenges such as high costs and low data volume due to the long tail problem
01
Industrial-level high-performance AI models faces challenges such as high costs and low data volume due to the long tail problem

Designing and training industrial high performance AI models requires a lot of cost investment and deep technical precipitation, and the long tail problem can hardly train high performance AI models because of low frequency and low data volume.

Demands for large-scale AI infrastructures emerge in specific industries first
02
Demands for large-scale AI infrastructures emerge in specific industries first

With the penetration and application of AI technology, demands for large-scale and diversified AI applications have grown rapidly in some industries, and demands for application implementation have gradually emerged in more potential industries.

Exponential growth of data volume and models puts forward high requirements for computing power
03
Exponential growth of data volume and models puts forward high requirements for computing power

With the evolution of AI models, their complexity and model parameters have been increasing, and the computational power required for training models has increased exponentially.

Concept positioning

Build a new type of AI infrastructure with high efficiency, low cost and scale, and comprehensively build a physical space digital search engine and recommendation system

×
Core architecture

SenseCore consists of three parts: the model layer, the deep learning platform, and the computing infrastructure

  • Model Layer
  • Deep Learning Platform
  • Computing Infrastructure
Works together with the SenseTime Joint Laboratory of the Chinese University of Hong Kong to build  OpenMMLab, an algorithm open source program, and cooperates with the Shanghai Artificial Intelligence Innovation Center to launch OpenDILab, an open source platform for decision-making intelligence, jointly creating an innovation ecosystem.
Model Factory
Model Factory
Has launched
46,000commercial AI models
OpenDILab Open Source Platform
OpenDILab Open Source Platform
Has launched
60 general decision-making AI algorithm series
OpenMMLab Open Source Framework
OpenMMLab Open Source Framework
GitHub Over
More than60,000stars
An industrial AI production open platform integrating large-scale AI computing power management, professional AI tool chain, and open AI algorithm, realizes the whole link and batch process from data annotation, algorithm design, to model training and deployment.
Model Compression Tool
Model Compression Tool
Relies on leading compression technology
Convert large models into lightweight models, and consumes less memory, while maintaining the same accuracy
Training Data Platform
Training Data Platform
Covers training data sets
All stages offull life cycle
SenseParrots Training Framework
SenseParrots Training Framework
SenseTime’s Leading Visual Algorithm Training Framework
Utilizes GPU cluster computing power efficiently, and achieves a 91.5%+ parallel efficiency on 1000 GPUswhen training a single large model, marking the leading level in the industry
Cross-platform Model Deployment Tool
Cross-platform Model Deployment Tool
Provides cross-platform compatibility
Improves reasoning efficiency by20%-80%, saves resources by15%-50%, and provides at least 500 millionusers with AI reasoning services
Located in Lingang, Shanghai, SenseTime Artificial Intelligence Data Center (hereinafter referred to as “SenseTime AIDC”) is an open, large-scale, low-carbon and energy-saving advanced computing infrastructure, which supports cloud-based all-round AI model production and deployment services. It is expected to generate a total computing power of 3.74×10^18 floating point operations per second, and can complete training of the model which has the most parameters in the world in one day. It is currently one of the largest AIcomputing centers in Asia.
  • AIDC
    Total Computing Power
    3.74×10^18×10^18floating point operations per second
  • AI Chips and Edge Devices
    Supported Large Vision Field Model
    10 BillionParameters
  • Sensors and ISP Chips
    Complete Training Done in 1 Day
    10 billion-levelParameter Model
Computing Infrastructure
Industry value
  • Accelerates the large-scale implementation of scale
    Accelerates the large-scale implementation of scale

    Accelerates the large-scale implementation of AI, and reduces the production costs and technical threshold of algorithm models.

  • Solves the long tail application problems
    Solves the long tail application problems

    Comprehensively solves the long tail application problem in urban management, enterprise service and personal life, and opens up the closed loop of business value.

  • Builds a new business mode
    Builds a new business mode

    Structures unstructured data, establishes a digital search engine and recommendation system based on physical space, and builds a new business model.

Inclusive AI creates a nice future
Inclusive AI creates a nice future Inclusive AI creates a nice future
Continuously update the whole line of products and insist on sincere communication and win-win cooperation

Help you achieve new breakthroughs in business with professional AI solutions and advanced AI products