On July 6, 2023, Nature, a top academic journal, published the latest research results of HUAWEI CLOUD's Pangu Meteorological Model, which is the first Nature-based paper published by a Chinese technology company as the only signatory, which has aroused widespread attention from the artificial intelligence industry at home and abroad. The Pangea Meteorological Model has achieved a major breakthrough in the field of weather forecasting by artificial intelligence, which can increase the prediction speed by 10,000 times compared with traditional methods, and can provide global weather forecasting in seconds.
The artificial intelligence big model is a hot new track in the field of science and technology, and it is also a highland of scientific and technological innovation that countries are vying to seize. According to the "Chinese Intelligent Big Model Map Research Report" released by the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology, the number of large models developed in the mainland ranks second in the world.
The industry view is that large models must not only be able to write poetry and draw, but also need to be able to combine with the real industry and serve thousands of industries, which is also the general direction of Chinese technology companies to break through in the new round of artificial intelligence.
Under the "100-model war", AI has penetrated into thousands of industries into a breakthrough
According to the statistics of the "Research Report on the Map of Chinese Labor Intelligence Big Model", 79 large models with more than 1 billion parameters in the mainland have been released, and domestic artificial intelligence large models have entered the era of "100-model war". Deeply cultivating the industry and using large models to accelerate the digital and intelligent transformation of thousands of industries has become an industry consensus and a breakthrough in the large model track.
In recent years, the combination of industry knowledge and big model capabilities is bringing significant changes to more and more industries. Taking the Pangu model released by HUAWEI CLOUD in 2021 as an example, many industry models and capability sets have been launched for mining, pharmaceutical molecules, meteorology, government affairs, finance, manufacturing, and railways, bringing great changes to many traditional industries.
In the field of meteorological science, previous AI weather forecasting methods faced challenges such as low accuracy, poor interpretability, and inaccurate prediction of extreme weather. HUAWEI CLOUD Pangea Weather Model is the first AI prediction model with more accuracy than traditional numerical prediction methods, which uses 3D high-resolution AI weather forecasting methods to complete global weather forecasts from one hour to seven days in seconds, including potential, humidity, wind speed, temperature, and sea level pressure, and the prediction speed is 10,000 times faster than traditional numerical methods.
In May 2023, the Pangea Meteorological Model forecast the steering path of the wind "Mawa" five days in advance. In August 2022, the Pangu Meteorological Model predicted the trajectory and landing time of Typhoon Saddle with an accuracy rate of 90%, far exceeding the industry average.
In the field of drug research and development, artificial intelligence big models have broken the "double 10 rule" of the past decades (the average cost of new drug research and development exceeds 1 billion US dollars, and the research and development cycle is more than 10 years), greatly accelerating the process of new drug development and marketing, and benefiting human health.
HUAWEI CLOUD Pangu Drug Molecule Model generates 100 million new small molecule compounds by learning 1.7 billion compound molecules in nature, shortening the lead drug development cycle from several years to one month and reducing R&D costs by 70%. Based on the large molecular model of Pangu drugs, the team of Professor Liu Bing of the First Affiliated Hospital of Xi'an Jiaotong University has developed a super antibacterial drug Drug X (cinnamoylmycin), which is a broad-spectrum antibiotic that can fight super-resistant bacteria, and is also the world's first new target and new class of antibiotics in the past 40 years.
In the coal mining industry, big models are changing the face of thousands of years of toil and dangerous underground operations, allowing more coal miners to work in suits and ties. The Pangu Mine Model can cover more than 1,000 subdivision scenarios such as mining, mining, machinery, transportation, and transportation of coal mines, helping coal mining enterprises build a safe and efficient intelligent system with fewer people and no people. At present, the Pangu Mine Model has been used on a large scale in 8 mines across the country.
In the rail sector, big models are making rail freight safer. In the past, the safety detection of truck tracks relied on artificial visual fault identification, which was not only labor-intensive, but also easy to cause fatigue and cause misjudgment. The Pangu Railway model can accurately identify 67 kinds of freight cars and more than 430 kinds of faults in the existing network, and the screening rate of trouble-free pictures is as high as 95%, which can reduce the labor intensity of workers by 95.75% and improve the operation safety of freight cars.
China's development of large models requires localized computing power and root technology innovation
In the era of large models, the growth of demand for computing power in all walks of life is much greater than the actual supply of computing power. Moreover, under the complex international competition pattern, the supply of advanced computing power is seriously insufficient, which has become the primary challenge for the development of domestic large models. This has once again sounded the alarm for China's science and technology community: the development of China's large models requires comprehensive independent innovation of localized computing power and AI root technology.
At the Huawei Developer Conference 2023 (Cloud) on July 7, HUAWEI CLOUD released Pangu Model 3.0 and Ascend AI Cloud Service, continuing to deepen its AI for Industries strategy and build another pole of global AI around Ascend AI Cloud Service. In terms of the design concept and development mode of the large model, the combination of a series of AI root technologies, such as the HUAWEI CLOUD Pangu large model and the Ascend AI cloud service, explores a path for the large model to be implemented faster.
In terms of pattern design, Pangu Big Model has pioneered a hierarchical decoupling design in the industry, and has formed a complete architecture including L0 basic large model, L1 industry large model, L2 industry subdivision scene large model and consistent capability set, creating a fast way for large models to penetrate into the industry. For example, in terms of industry capabilities, HUAWEI CLOUD can provide general large models for industries such as government affairs, finance, manufacturing, mining, and meteorology, and can also train proprietary large models for customers based on their own data.
In addition to large models, how to use software platforms and cloud services to lower the threshold of model development and improve the efficiency of large model training, inference, and deployment has become the focus of attention of the industry. Many AI root technologies, represented by Ascend AI Cloud Services, are the other pole of AI that HUAWEI CLOUD hopes to build in the era of big models. Through the integrated software and hardware platform architecture and rich cloud services, Ascend AI cloud services can make large model training and inference more efficient and stable, and improve large-scale parallel training capabilities.
For example, in terms of computing power, HUAWEI CLOUD empowers AI computing resources through cloud services, and builds an AI computing power cloud platform based on Kunpeng and Ascend. Currently, the Ascend AI cloud service with single-cluster 2000P Flops computing power has been launched in HUAWEI CLOUD Ulanqab and Gui'an AI Computing Power Centers. It allows enterprises to use abundant AI computing power at low cost, high efficiency, and on-demand to quickly iterate their own large models just like using cloud services.
In addition, AI root technologies such as CANN, a heterogeneous computing architecture under HUAWEI CLOUD, MindSpore, an all-scenario AI framework, and ModelArts, an AI development production line, provide key capabilities such as distributed parallel acceleration, operator and compilation optimization, and cluster-level communication optimization for the development and operation of large models. Based on Huawei's AI root technology, the training efficiency of large models can be tuned to 1.1 times that of mainstream GPUs in the industry.
Commenting on the development direction of the Pangu model, Zhang Pingan, executive director of Huawei and CEO of HUAWEI CLOUD, said, "The Pangu model should allow every industry, enterprise, and everyone to have their own expert assistant to make work more efficient and easier. We have always adhered to the strategy of AI for Industries and continued to move forward on the road of deep cultivation of the industry. Pangu was born for the industry, and I firmly believe that the big model will reshape thousands of industries. ”
END