Huawei Cloud provides out-of-the-box computing power for large model training and inference in Hong Kong

Huawei Cloud successfully held the Huawei Cloud Summit on April 23, providing ready-to-use AI cloud services in Hong Kong, providing efficient, long-term, and reliable AI computing power for large model training and inference.

Huawei Cloud stated that its infrastructure will support efficient migration, development, and efficient operation of large models through a full-link cloud tool chain, and will provide a large model area specially optimized for Ascend Cloud to enable "hundred-digit models". The "Module Thousand Status" application was quickly implemented.

Shi Jilin, President of Huawei Cloud Global Marketing and Sales Services, was at the event.

Shi Jilin, President of Global Marketing and Sales Services of Huawei Cloud, said: Hong Kong has a good foundation for the development of AI industry, with first-class university resources and scientific research institutions, as well as an open economic system and The international business environment attracts top talents and companies from around the world to gather here. Huawei Cloud supports Hong Kong in establishing a complete innovative technology ecosystem and turning Hong Kong into a global AI industry highland. In order to better support the "hundreds of modes and thousands of states" innovation of large models, Huawei Cloud has built a full-stack independent, innovative AI cloud computing power base that combines software and hardware in Hong Kong and multiple cities in the mainland to provide large-scale, low-cost, high-efficiency, Easy-to-use smart computing power.

Huawei Cloud officially released the Shengteng AI cloud service in 2023, relying on the three major AI cloud computing power centers in Gui'an, Ulanqab, Wuhu and Hong Kong nodes to provide surging computing power for large model training. Huawei's Ascend AI cloud service supports high efficiency and long-term stability, providing longer and more stable AI computing power services. It can realize large cluster training without interruption for 30 days. Training operation faults can be discovered in 1 minute, diagnosed in 5 minutes, and restored in 10 minutes.

Previously, iFlytek provided Ascend clusters from Huawei for iFlytek Spark large model training, and the overall training performance improved by 17%. After the release of Spark 3.0, faced with the new challenge of insufficient inference resources, iFlytek adopted Huawei Cloud Ascend AI cloud service. On the one hand, it can quickly activate hundreds of cards of inference resources. On the other hand, it can adjust the resource usage at any time according to the business launch situation. No investment will be wasted. At the same time, Huawei Cloud's global computing power layout also supports iFlytek's business overseas and jointly serves the intelligence of global enterprises.

"Meitu's self-developed AI visual large-scale model MiracleVision (Qixiang Intelligence) is widely used in the five major industries of e-commerce, advertising, games, animation, and film and television, helping to improve the efficiency of designing scenarios in subdivided fields. In Huawei Cloud With the help of Shengteng AI Cloud Service, the models used in scenarios such as text generation and graph generation were migrated to Shengteng AI Cloud Service. Both parties jointly optimized more than 30 operators and accelerated the process in parallel. After the migration, , Meitu AI painting and other business reasoning increased by 30%, helping enterprises to better achieve cost reduction and efficiency improvement.”

Huawei Cloud cooperates with software partners to build joint solutions and provides them in Hong Kong based on Shengteng Cloud services. The AI ​​computing power and cross-border trusted network and compliance framework carry multiple sub-functions such as intelligent document analysis, commercial big data and compliance auditing, and risk control management, and can provide enterprises with AI assistance such as qualification verification, classification, and identification. capabilities, speed up the review and approval of fund applications, optimize end-to-end process services, and significantly reduce related manpower investment.

According to our understanding, on the cloud, Huawei Cloud also provides Pangu large model 100 capability set 70 open source large model adapted and optimized based on Ascend Cloud, as well as a full-process model development and tuning tool chain to provide local enterprises with Provide full-stack computing power services, and quickly implement "hundreds of models and thousands of states" applications through solid cloud computing power base capabilities.

The above is the detailed content of Huawei Cloud provides out-of-the-box computing power for large model training and inference in Hong Kong. For more information, please follow other related articles on the PHP Chinese website!