The team led by Prof Shang Jiandong at the National Supercomputing Center at Zhengzhou University, in collaboration with the Henan Academy of Geology, launched "Kunshu", a geology-focused large language model, on July 8. This model, the first of its kind in China, integrates domestic high-performance computing and artificial intelligence, marking a significant breakthrough in the fusion of geological information technology and domestic technology by Zhengzhou's supercomputing capabilities.
"Kunshu" addresses challenges in geological surveys via a novel task-driven grouped mixture of LoRA experts (TG-MoLE) architecture, providing precise and efficient solutions for complex geological tasks. The deployment of this model strengthens the digital foundation of the geological industry, ensuring national energy and resource information security while advancing various geological tasks and pioneering a new paradigm for mineral exploration research.
The model not only enhances geological science literacy but also offers intelligent services to the public. Users can now experience the model's capabilities through the National Supercomputing Center. Moving forward, the team will continue optimizing the model's performance and expanding its applications in deep earth exploration and space-to-earth observations, injecting new momentum into the high-quality development of the geological industry in the new era.