Dr Zhu Shenggao, Vice President of AI, Huawei Cloud Middle East & Central Asia discusses the theme of GITEX Global 2023 and what sets them apart from their previous participation at the event.
- What’s the theme of you participation at GITEX this year? What are the key aspects that you’re highlighting at the show?
This year, on company level, the theme for GITEX is “Accelerate Intelligence”, and Huawei Cloud theme is “Everything as a Service, Pangu Model Reshaping AI for Industries”. Large models and related applications are the most complex hardware and software system engineering, which requires systematic innovation. Huawei Cloud is committed to building the cloud foundation for an intelligent world with systematic innovation. Huawei Cloud innovates in fields such as Pangu large models, application modernization, AI development pipeline, and digital content production pipeline. This year at GITEX, Huawei Cloud is sticking to the “AI for Industries” strategy, and will demonstrate how the 5+N+X architecture of Pangu enables intelligent upgrade of industries such as government, electric power, oil and gas, and meteorology. Huawei Cloud will also reveal its Huawei Cloud Stack 8.3.0 – a hybrid cloud platform, which can be used to deploy AI cloud services and key products of large models at on-premises data centers of enterprises. With robust capabilities and extensive expertise of Huawei Cloud Stack and Pangu Models, we hope to empower every enterprise to build their own large models.
- What differentiates your participation this year as compared to previous years?
Huawei continues to bring the latest innovation and showcase the most innovative products and cloud services. This year, we are focusing on Accelerating Intelligence, to build the cloud foundation for an intelligent world and reshaping industries with AI. Huawei Cloud will showcase how the 5+N+X architecture enables intelligent upgrade of industries such as government, electric power, oil and gas, and meteorology, and empower every enterprise to build their own large models through the robust capabilities and extensive expertise of Huawei Cloud Stack and Pangu Models. The Pangu Model 3.0 is the first commercial Arabic model trained with hundreds of billions of parameters, it is deployed on public cloud and on-premises allowing customers to build their own models in the way they prefer. The decoupled and hierarchical architecture allows the Pangu model to be quickly adapted to a wide range of downstream tasks. Customers can load independent datasets to train their own models. To meet enterprises’ data security requirements for AI models, Huawei Cloud provides multiple deployment modes, including public cloud, dedicated zone (public cloud), and hybrid cloud. In this way, enterprises can enjoy the continuous supply of compute resources from the public cloud while keeping data within their own organizations to build their own AI models.
- What does ‘AI Everything’ mean to you?
AI Everything mean to Huawei Cloud is AI for Industries. Huawei Cloud has been devoted to “AI for Industries”, and that’s what Pangu models are designed for. By tackling big challenges head-on, our Pangu models will unlock more possibilities for all industries.
- What are the challenges and opportunities in ‘AI Everything’? How do you aim to support organisations in this journey?
AI has entered into core production systems and has started creating greater value, and AI adoption is expected to reach 20% by 2026 in major industries. Large models are behind the latest AI explosion. When the model size grows above a certain threshold, the model accuracy suddenly rises to a very high level, or even above human level.
Huawei has always practiced AI in different Industries, enabling the intelligent upgrade of thousands of industries. To operationalize large model across different industries, there are three challenges:
- Large models are highly versatile but weakly specialized. The specific scenarios need to have professional knowledge, understand the industry workflow, and give professional and accurate answers.
- Large models contain a lot of basic knowledge, but the skills to process it are insufficient. At present, the general-purpose large models are mostly general-purpose language models. However, the models need to have specific skill to process specific task as the enterprise scene is complex.
- Industry data is one of the core assets of an enterprise. We need to ensure the security and compliance of enterprise data.
Huawei Cloud is committed to reshape AI for Industries. With more than 1,000 AI projects and we have always adhered to the “AI for Industries” strategy.
Pangu models were born to serve the needs of industries. A three-layer, decoupled architecture allows Pangu models to be adapted to a wide range of downstream tasks more easily.
The L0 layer of Huawei Cloud Pangu models provides over a hundred preset capabilities. Our enterprise customers can select datasets based on their needs and use preset Pangu workflows to quickly develop new capabilities for industry models.
The L1 layer consists of a number of industry-tailored models. We can provide customers with industry models we have pre-trained on open industry datasets. Alternatively, customers can train their own models using their own datasets based on Huawei’s Pangu models.
The L2 layer provides pre-trained models for specific industry scenarios and tasks. Customers can use built-in fine-tuning workflows provided by ModelArts to fine-tune models based on scenario-specific needs and deploy them as AI applications. This helps to simplify AI development and improve efficiency.
We are committed to helping industry customers to train and use large models in a secure and efficient manner.
Our Pangu models are industry-tailored with specific knowledge and experiences, such as our own business experience over the past 30 years, know-how in more than 10 industries, and those from our customers and partners. In addition to an ocean of general knowledge, Pangu models have been pre-trained with open datasets from more than 10 industries, including finance, government, meteorology, healthcare, health, Internet, education, automotive, and retail. The data volume of each industry exceeds 50 billion tokens. Pangu models have been deployed by more than 10 industries and sectors, such as government, finance, mining, manufacturing, and healthcare, and have started to create solid value for our customers.
We are committed to empower enterprises from different industries in the region to build their own large models.