Shulou(Shulou.com)11/24 Report--

Today, large model technology continues to infiltrate into the major scenarios of the industry. The key to empowering large models is to strengthen the construction of high-performance networks. Du Lesser Fullness of Grain not only innovated the large model technology, carefully developed the "Xuanyuan" big model, but also jointly with Baidu Smart Cloud launched the "Intelligent Computing Center Network Architecture White Paper" to make high-performance network construction simple and intelligent.

Lesser Fullness of Grain officially opened up the first 100 billion parameter model in China.

On September 5, at Baidu Yunzhi Conference, Baidu Intelligent Cloud Joint degree Lesser Fullness of Grain launched the "White Paper on the Network Architecture of Intelligent Computing Center." The white paper, written by Baidu Intelligent Cloud and Lesser Fullness of Grain, a front-line expert responsible for high-performance network landing, covers all aspects of high-performance network construction. it includes the architecture selection of clusters of different sizes, the construction of visual operation and maintenance tools on cloud platforms, and the operation plan design of multi-tenants, so as to provide full-process guidance for the industry and make complex high-performance network construction simple.

Large models need big computing power, and how to build a high-performance network is one of the most critical steps. Lesser Fullness of Grain officially opened up China's first 100 billion parameter model-"Xuanyuan" in May this year. In the task evaluation in the scene, Xuanyuan surpassed the mainstream open source model in the market and won 63.33% of the 150 responses. With the rapid development of large models, the demand for underlying computing power of the model is increasing exponentially. The continuous expansion of the scale of intelligent computing power brings the increasing expansion of the scale of GPU computing power deployment and the increasing requirements for high-performance networks.

Degree Lesser Fullness of Grain provides stable network guarantee for the training of "Xuanyuan" big model.

As the parameter scale of the degree Lesser Fullness of Grain model exceeds 100 billion, in order to meet the increasing demand for computing power, the intelligence center of the Lesser Fullness of Grain private cloud model is also constantly upgrading. The large-scale computing network with low delay, large bandwidth, stable operation and high maintainability is an important cornerstone of the intelligence center. In the process of building a private cloud intelligent computing center, du Lesser Fullness of Grain combined with Baidu Intelligent Cloud to build a set of private cloud intelligent computing network. While supporting the scale of Vanka GPU, compared with the common Dragonfly and Tours network topologies in the industry, the network bandwidth is more sufficient, the number of hop steps between nodes is more stable, the network delay index is shortened to less than 3 microseconds, and it has end-to-end observability and fault self-healing ability. The non-blocking, low-delay and highly reliable network design effectively supports the rapid iteration and development of upper-level intelligent computing applications, and provides a stable network guarantee for the training of the "Xuanyuan" large model.

At present, the Lesser Fullness of Grain Intelligent Computing Network Center has been built into one of the first Vanka-scale intelligence centers in the domestic industry under the private cloud model, providing a stable network guarantee for the training of the "Xuanyuan" model.

In view of the large model, du Lesser Fullness of Grain strengthened the investment in research and development, constantly conquered the technological fortress, developed new technologies and achievements, and accelerated the application of the large model in various fields, while the release of the White Paper on the Network Architecture of the Intelligent Computing Center also provided a sample for the industry to study the large model technology. Next, du Lesser Fullness of Grain will continue to aim at cutting-edge technologies such as big models, so that innovative technology will become a "new engine" for the development of the industry.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.