Global First Billion-Parameter Seismic Wave Model "Diti" Releases New Version

近日，全球首个拥有超过一亿参数的地震波大型模型——“谛听”，在国家超级计算机成都中心推出了其第三阶段测试版。这个新版本成功分析了西藏定日发生的6.8级地震数据，并通过实际应用证明了其效能。

据透露，“谛听”地震波大型模型计划于2025年向公众开放其微调、推理框架及相关的数据处理流程。届时，用户将能够利用这一模型在超算平台上开展业务分析和科学研究。

根据报道，中国地震局地球物理研究所使用“谛听”地震波大型模型对2024年12月7日至2025年1月7日期间，在西藏定日6.8级地震震中附近的珠峰站和日喀则站记录到的地震连续波形进行了分析。该模型自动检测到了震前事件共452次，并在震后27小时内识别出429次余震事件。此外，该模型还生成了基于单台定位的AI地震序列目录。

“谛听”地震波大型模型是由国家超级计算成都中心、中国地震局地球物理研究所与清华大学共同开发，并于去年七月正式推出。短期内，该模型将主要用于地震信号识别、地震活动监测以及大地震快速响应等方面。长远来看，它有望深化地震学领域对观测数据的理解，促进地震学研究的进步。

Recently, the world's first large-scale seismic wave model with over 100 million parameters - "Diti", launched its third phase test version at the National Supercomputing Chengdu Center. This new version successfully analyzed data from a 6.8 magnitude earthquake in Dingri, Tibet, and demonstrated its effectiveness through practical application. It is reported that the "Diti" large-scale seismic wave model plans to open up its fine-tuning, inference framework, and related data processing procedures to the public in 2025. At that time, users will be able to utilize this model on supercomputing platforms for business analysis and scientific research. According to reports, the Institute of Geophysics under the China Earthquake Administration used the "Diti" large-scale seismic wave model to analyze continuous waveform data recorded by stations near the epicenter of the 6.8 magnitude earthquake in Dingri, Tibet, between December 7, 2024, and January 7, 2025. The model automatically detected a total of 452 pre-earthquake events and identified 429 aftershock events within 27 hours after the quake. Additionally, the model generated an AI-based seismic sequence catalog based on single-station positioning. The "Diti" large-scale seismic wave model was jointly developed by the National Supercomputing Chengdu Center, the Institute of Geophysics under the China Earthquake Administration, and Tsinghua University, and officially launched last July. In the short term, the model will mainly be used for seismic signal identification, earthquake activity monitoring, and rapid response to major earthquakes. Looking ahead, it is expected to deepen the understanding of observational data in seismology and promote further development in earthquake studies.