DiTing, the world's first professional seismic data processing model with 100 million parameters, was released in Chengdu, Sichuan province southwest China, in late July, and will be publicly tested by the end of 2024. The one billion parameter version is expected to complete pre-training in August 2024.

As an AI model, DiTing is based on one of the biggest data sets in the world and can be used to spot earthquake signals, monitor seismic activity and support rapid responses to earthquakes, according to its developers.

The model was developed by the National Supercomputing Center in Chengdu in cooperation with the Institute of Geophysics of the China Earthquake Administration and Tsinghua University, Xinhua reported.

Guo Li, chairman of Chengdu Supercomputing Center Operation Company, said the model with 100 million parameters can already be used to make earthquake predictions.

Chen Shi, deputy director of the Institute of Geophysics, China Earthquake Administration, said, "Based on the massive data of China earthquake observation network, the large AI model has significantly improved the accuracy and speed of seismic signal recognition." Seismology is an observational science and breakthroughs often come from a deep understanding of observational data.

Wang Jianbo, executive deputy director of the National Supercomputing Center in Chengdu, said the creation of DiTing is of great significance for breaking the performance bottleneck of small and medium seismic wave models. This can improve the ability to process seismic big data intelligently and the level of information mining.

The computing power provided by the Center is the basis for developing the large seismic wave model.

Compared with the small seismic wave model, the training data and the number of parameters of the large model have increased significantly, by more than 100 times, leading to a substantial increase in the demand for computing power.

To this end, the Center provides key technical support for optimizing code efficiency, making the R&D process of? DiTing smoother and more efficient, Wang said.

In the future, the large model could be applied in fields such as mine seismic monitoring, shale gas exploitation, urban underground space structure detection, underwater earthquake monitoring and others, according to the developers.