Huawei Cloud introduces Pangu Big Model 5.5

On June 20, Huawei Cloud launched the Pangu Big Model 5.5. The new AI technology comes with various upgrades in terms of understanding and processing. At the same time, it can deliver better responses and an improved user experience.

Pangu refers to Huawei’s AI LLM (Large Language Model). This time, the company has upgraded five basic models with Pangu 5.5. These are: NLP – natural language processing, computer vision, multi-modal, prediction, and scientific computing.

Zhang Pingan, the CEO of Huawei Cloud, launched the new Pangu Big Model 5.5. He also introduced the major changes in the sub-models of this technology.

Let’s check them one by one!

NLP: Pangu 5.5 NLP has over 256 experts to enhance reasoning, tool calling, mathematics, and more. It has been trained by the Ascend Cloud architecture. Pangu 5.5 adopts a combination of fast and slow thinking, as well as two-stage progressive training.

Thus, the model can think according to the difficulty of the problem. It can quickly respond to simple problems while thinking deeply about the complex ones. Huawei has increased the reasoning efficiency of Pangu 5.5 by 8 times. It can answer 10 jumps of complex questions in 5 minutes and generate a 10,000-word report.

Computer Vision: Huawei has introduced a 30-billion parameter CV model with a new MoE architecture – the largest visual model in the industry. It supports multi-dimensional, pan-visual perception, analysis, and decision-making capabilities.

The Pangu CV model can also build a scarce pan-visual for industries like oil and gas, transportation, coal mines, and more.

Huawei Pangu Big Model 5.5 capabilities (Image Credits: Weibo)

Multi-Modal: Pangu 5.5 multi-modal can easily develop the required digital physical space for the training of smart driving solutions and robots. It can generate driving video of each camera without depending on the expensive road sampling.

Prediction: Huawei Pangu Big 5.5 Prediction Model uses the industry’s first unified pre-training framework of triple transformer. Data like time series of equipment operation logs, image data, and tabular data can be uniformly triple-encoded and efficiently processed within the same framework.

This overall mechanism improves the accuracy of large prediction models and boosts generalization across industries and related scenarios.

Scientific Computing: Huawei has integrated AI into the Pangu scientific computing model. As a result, it will reduce the error of a single forecasting model. The tech will improve the forecast and warning abilities of a disaster weather.