According to China’s Cheil Industries and others, Huawei introduced an industrial business AI model version 3.0 that can be used in various industries such as finance, administration, manufacturing, and pharmaceuticals at the “Huawei Developer Conference 2023” held in Dongguan, Guangdong Province, China on the 7th.
According to Huawei Cloud, Pangu 3.0 has up to 100 billion parameter functions. It allows various customers to train AI models according to their needs. Pangu 3.0 provides more personalization functions and safety to users compared to Chat GPT and Bing, which are AI models in the United States. It provides solutions optimized for the Chinese language environment by enhancing its performance in the field of Chinese natural language processing (NLP). Zhang Pingan, Huawei’s executive director and Huawei’s cloud CEO, said Pangu 3.0 will “reorganize thousands of industries.”Baidu and Alibaba have entered the Chinese version of Chat GPT market, but Huawei has cited examples in some areas, emphasizing that unlike them, it is an optimized model for industrial business markets.A typical field is weather forecasting models. Huawei explained that Pangu Weather is the first AI prediction model that shows higher precision than traditional numerical weather forecasting methods. He stressed that using this model can improve the prediction speed by 10,000 times, reducing the global weather forecast time to just a few seconds. On the 5th, the international academic journal “Nature” also published a paper on Huawei’s plate weather.According to Huawei, four to five hours of simulation was needed to predict the trajectory of the typhoon over the next 10 days in a high-performance cluster of 3,000 servers in the past, which Pangu Weather claimed could be processed in 10 seconds on a single graphics processing unit (GPU) of a single server.The plate mining model increased the operational stability of underground coal mines.
It has the advantage of being able to handle 1000 sub-works for the seven major work processes of the coal mine and monitoring the underground situation from the ground.In the railway sector, 67 types of trucks and more than 430 types of failures can be accurately identified. The accuracy is 99.8 per cent. Assuming that an average of 1,000 trains pass through the station every day, 260 train inspectors had to visually check 4 million images every day. Pangu railway models can quickly scan all images and filter 3.8 million defect-free images. It can process 95% of the entire image. Huawei explained that this greatly improves efficiency as train inspectors only need to process the remaining 200,000 images.In addition to these fields, R&D costs can be reduced by 70% when used in the field of drug molecular models. In the case of manufacturing, it helps to increase efficiency and save money. In the past, three hours were needed to make a one-day plan, but with the Pangu model, it is possible to make a three-day plan in just one minute.Pangu 3.0 is also characterized by training and learning through Huawei’s own AI processor “Sungtung (English name Ascend)” and AI computing framework “Sungs” (English name MindSpore) instead of GPUs that are mainly used in the industry. CEO Jang said, “As Chinese companies have to spend high prices and a long time to secure NVIDIA GPUs, it is difficult to keep up with the ever-increasing demand for computing power,” adding, “We hope Huawei will provide an alternative to computing power.”
KS CHOI
US ASIA JOURNAL
