As foundation model technology continues to advance toward domain customization,multimodal integration,and multitask processing,Telecom large models are demonstrating unprecedented potential and value in enabling network intelligence transformation,optimizing resource allocation efficiently,and enhancing user experience.This paper systematically reviews the taxonomy and primary network functions of Telecom large models from the perspectives of cloud,edge,and terminal layers.Furthermore,it explores pretraining and parameter fine-tuning strategies for these models,with particular emphasis on task-oriented fine-tuning and interactive collaborative inference techniques.Through the analysis of successful practical cases,such as China Mobile's Jiutian Model,China Unicom's Yuanjing Model,and China Telecom's Xingchen Model,this study delves into the core advantages and significant potential of Telecom large models in refined network management,efficient data transmission strategy formulation,and comprehensive user experience enhancement.Finally,the paper envisions the broad prospects of Telecom large models in advancing future network intelligence,exploring innovative service paradigms,and improving operational efficiency.It also highlights key technological challenges,including stringent guarantees for data privacy and security,enhancement of model generalization and robustness,optimization of real-time processing capabilities,and improvement of system scalability.
communication networkslarge modelsnetwork intelligencemodel optimization
维普
万方数据
