Huawei recently held the Huawei Cloud Industry Summit Forum 2023 in Beijing. The theme of the summit was “Practice Deep Use of the Cloud and Accelerate Intelligent Upgrades.” At the meeting, Huawei Cloud launched the industry’s first large-model hybrid cloud. Huawei Cloud Stack 8.3 is the first in the industry to implement large model capabilities based on hybrid cloud deployment. It offers the industry’s most complete AI production chain such as a computing power platform. It also offers cloud services, development kits, and professional services. This tech will help government and enterprise customers build an exclusive large model in one stop. Huawei Cloud Stack has made four major optimizations for large model scenarios
Gizchina News of the week
Hybrid cloud optimizations
- Based on the native hybrid cloud capabilities, users can extend large models from local to the edge and public cloud. The multi-cloud collaborative architecture allows large industry models to get training on the public cloud. They can be fine-tuned on the hybrid cloud based on local enterprise data and then pushed to the edge cloud.
- Through the collaboration of software and hardware, operator fusion and mixed precision optimization are realized. Huawei said that the model training performance has improved by 45%.
- Huawei Cloud Stack provides three types of tool suites. It allows enterprises to efficiently complete data cleaning, model development, and application development, and lower the development threshold.
- Large model training tasks usually require continuous execution for several days. Huawei Cloud Stack allows AI training to run stably for more than 30 days through the ability to resume training without any sense of breakpoints.
Huawei Cloud Stack has teamed up with more than 10 government and enterprise customers to explore large industry models based on hybrid cloud. Huawei Cloud also released the “In-depth Cloud Usage Outlook 2025” white paper. The paper focuses on the next three years and looks forward to the top ten techs and models that will be widely implemented in the government and enterprise markets. It hopes to work with the industry to help more companies move towards in-depth cloud usage.
