the "magic machine model" was born. the sun yat-sen university team released the first ai model in the field of emergency science popularization | science station
2024-09-24
한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina
the reporter learned from the school of intelligent engineering of sun yat-sen university that the first multimodal large language model specially designed for emergency science popularization, the "shenji large model", has been completed by the team of shen ying of the school and officially released recently. the large model deeply integrates artificial intelligence technology with the emergency management system and has become an important tool for the popularization of emergency knowledge.
it is reported that the "shenji big model" can provide the general public with practical emergency knowledge and methods to deal with emergencies, help improve the public's safety awareness and self-rescue and mutual rescue capabilities, and open a new chapter in the popularization of emergency science for all. at the same time, in response to the pain points in the industry, the big model can provide governments, enterprises, institutions and the public with intelligent, efficient and humanized emergency scenario solutions.
according to reports, in order to enable the "shenji big model" to provide intelligent, humanized, and refined solutions in emergency science popularization, the laboratory team designed consultation, decision support, and emotional support question-and-answer tasks for multiple scenarios such as pre-prevention, in-process handling, and post-disaster recovery. a total of 205,617 instruction pairs were built by themselves, and 22,353 public data corpora from guidelines, white papers, news, real cases, search engines, online services, and multiple rounds of questions and answers were introduced for fine-tuning, further enhancing the model's generalization ability and response speed. the comprehensive performance evaluation system for the emergency science popularization big model proposed by the team covers the objective evaluation of safety emergency management knowledge questions and answers and the subjective evaluation of experts, ensuring the high standards of the model in terms of safety, practicality, and standardization.
to ensure the excellent performance of the "shenji big model" in emergency science popularization, the "shenji big model" was pre-trained for 2 epochs and fully fine-tuned for 3 epochs on 36 40gb nvidia a100 gpus, training the model versions with 7b and 72b parameters respectively. in the evaluation of emergency management knowledge questions and answers, the model accuracy rate reached 71.53%, which greatly reduced the need for manual intervention and demonstrated excellent timeliness and reliability.
the model can provide more than 1,000 user responses within one minute, which is expected to reduce the burden on emergency personnel and improve emergency response efficiency. for governments, enterprises, institutions and ordinary people, the model provides refined emergency scenario consultation and emergency psychological counseling services to help build a safer and more harmonious social environment.
at present, the launched version 1.0 of the shenji big model has many practical functions: it can conduct questions and answers on common emergency safety knowledge, provide users with basic emergency consultation, respond quickly, and answer professional questions; based on retrieval enhanced generation technology (rag), the shenji big model traces the knowledge source of all generated answers, and provides users with reliable professional original knowledge document information with stronger professionalism; due to the different standards of various emergency management data, the shenji big model integrates massive historical data and rules and regulations for easy query, which solves the problems of knowledge fragmentation and data dispersion; the shenji big model can assist relevant practitioners in formulating emergency management plans, writing emergency plan documents, writing accident investigation reports, etc., greatly liberating labor costs.
in addition, after fires, construction accidents, earthquakes, etc., both the affected people and the rescue workers involved in the disaster relief may suffer from negative post-disaster psychological effects. the shenji large model can provide emergency psychological rescue and psychological counseling assistance.
shen ying, associate professor at the school of intelligent engineering of sun yat-sen university, said: "the release of the 'shenji big model' marks a solid step forward in our intelligent emergency science popularization. we hope that it can play an important role in the field of emergency science popularization and contribute to popularizing public emergency safety education and building a harmonious and safe society."
text｜reporter li gang guo siqi intern liu jiaqi
