Huawei's large model Pangu 3.0 has made its debut at the company's cloud developer conference. ZHANG Ping'an, CEO of Huawei Cloud, said the model consists of three layers of architecture: L0, L1 and L2, which will reshape thousands of industries.
The L0 layer comprises fundamental models for natural language, vision, multimodality, prediction, and scientific computing, catering to various industry scenarios. The L1 layer consists of industry-specific models, including government, finance, manufacturing, mining, and meteorology, which can be trained on public or proprietary data. The L2 layer offers specialized models for specific industry applications like government hotlines, branch assistants, and drug screening.
Pangu 3.0 provides a series of large language models with parameter counts ranging from 10 billion to 100 billion, offering new capabilities such as knowledge-based question-answering and image generation.
Pangu-Weather is the first AI to demonstrate greater accuracy than traditional numerical weather forecast, reducing global weather prediction time to just seconds.
Huawei also paraded new Ascend AI Cloud services, high-performance computing power and support for mainstream AI frameworks. The Pangu large language model is powered by Ascend.
Huawei’s investment in computational infrastructure includes AI centers in 25 cities, providing the market with a greater choice of computing power to address real-world problems.
