Consolidating the Intelligent Computing Base for Digital China

The Sea-mind Vision Large Model is an industry model jointly developed by China Telecom Cloud Technology Co., Ltd., E-Surfing Vision Technology Co., Ltd., and Digital Life Technology Co., Ltd.

E-Surfing Vision is the fifth basic network built by China Telecom to implement the "Digital China" strategy and carry out digital development. Currently, over 60 million devices have been connected to this network. Drawing on ubiquitous video terminal access, video scheduling, cloud storage, AI applications, and other capabilities, E-Surfing Vision serves many fields including homes, rural areas, communities, and industry governance.

Built on the intelligent computing base of eSurfing Cloud, the Sea-mind Vision Large Model upgrades the product system of E-Surfing Vision in all aspects. This Model is equipped with multiple core capabilities, including recognition of over 5,000 categories of tags, intelligent retrieval, analysis and comprehension, and visual super-resolution, through learning from hundreds of millions of training data and tens of billions of model parameters it possesses.

Industry-leading indicators prove that this Model can realize efficient computing, process various complex tasks, and enable four scenarios, including urban governance, industry vision, community and rural areas, and digital home, contributing to the blueprint of Digital China.

Take urban governance as an example. The Sea-mind Vision Large Model can efficiently analyze video streams, automatically generate tags, and support quick and accurate locating. This helps managers detect exceptions in cities in real time and prevent potential risks.

Another example is the "silver-haired economy" often mentioned on TV and newspapers. The increasing life expectancy of people develops the elderly care into an industry with huge space in China. The Sea-mind Vision Large Model will do a great job in this industry and create a new paradigm of home intelligence.

As a leading application model in the vision industry in China, the Sea-mind Vision Large Model offers an open and converged ecosystem. It uses innovative digital technologies and solutions to enable various industries, serve thousands of households, and bring digitalization to everyone to consolidate the foundation of "Digital China".

Eliminating the Computing Power Threshold of Foundation Models

The base of the Sea-mind Vision Large Model is the WisHub one-stop intelligent computing service platform. This Platform is an epitome of foundation model industry requirements and application implementation. Through systematic optimization of software and hardware, this Platform considers the usability, functionality, and security, cost, and other aspects for customers.

The WisHub Platform summarizes and extracts key processes, complex technologies, and valuable experience during model development and training to build a one-stop pipeline covering the entire life cycle and the whole process of model production and application. This platform significantly lowers the threshold for model training, fine-tuning, deployment, and inference, and enables customers to focus more on model upgrades and applications.

Based on four platforms on data, model development, model service, and application service, the WisHub Platform provides multiple product functional modules, such as data preparation, model development, task management, and model optimization. Functions and components can be flexibly assembled to help users build dedicated toolsets for different scenarios.

Computing platforms are showing more significance in the foundation model era. For organizations and enterprises, an efficient intelligent computing platform not only helps them cut costs, but also deploy models more quickly to seize market opportunities.

Lighting the Future of Industry Intelligence

As we know, the foundation model computing platform involves a wide range of aspects, including computing, network, storage, and framework. The utilization efficiency of any resource will affect the efficiency of the computing platform, which further affects the training progress and deployment efficiency of models.

In addition to the general capabilities of the computing power platform, continuous technological innovation and breakthroughs of eSurfing Cloud are the key factors for implementing model applications and accelerating AI convergence in more application scenarios and industries.

In view of this, the WisHub Platform builds reasonable and flexible task allocation and scheduling policies at the computing power scheduling layer, senses the underlying GPU, IB, and NVlink networks, and obtains the GPU and network status in real time, laying a foundation for task allocation and scheduling and plays a core role in connecting the upstream and downstream.

With these functions, the WisHub Platform can quickly reschedule resources for resumable training after the underlying hardware is faulty, securing the training results before fault occurrence. This supports the training task progress and ensures the overall investment of customers.

Besides, Until Now,  eSurfing Cloud  intelligent computing scale deployed in China has reached 4.7 Eflops.