文献类型：会议论文

英文题名：A Lightweight Generative Large Language Model (LLM)-based Internet of Things (IoT) Networking Framework for Edge-Cloud Collaboration

作者：Dai, Qinglong[1];Li, Ran[2];Qin, Guangjun[1];Yang, Zixuan[1];Liu, Guangnan[1]

第一作者：Dai, Qinglong

通讯作者：Dai, QL[1]

机构：[1]Beijing Union Univ, Smart City Coll, Beijing, Peoples R China;[2]Beijing Union Univ, Tourism Coll, Beijing, Peoples R China

第一机构：北京联合大学继续教育学院

通讯机构：[1]corresponding author), Beijing Union Univ, Smart City Coll, Beijing, Peoples R China.|[1141733]北京联合大学继续教育学院;[11417]北京联合大学;

会议论文集：2025 Workshop on the Edge-Cloud Collaboration for AI-CoNEXT

会议日期：DEC 01-04, 2025

会议地点：Hong Kong, PEOPLES R CHINA

语种：英文

外文关键词：LLMs; Generative IoT; Edge computing; Edge-cloud collaboration; Experiment

摘要：Large language models (LLMs) have emerged as promising enablers for the Internet of Things (IoT), owing to their advanced reasoning capabilities and extensive pre-trained domain knowledge. However, the substantial number of parameters in LLMs imposes significant computational and storage demands, typically necessitating highperformance hardware. These stringent resource requirements are inherently incompatible with the resource-constrained nature of edge environments. To address this challenge, we propose a Lightweight Generative LLM-based IoT (LGL-IoT) networking framework for edge-cloud collaboration. The framework comprises a hierarchical edge-cloud architecture and an autonomous response mechanism, enabling adaptability to dynamic IoT environments. A compact version of the LLM is deployed on local edge devices to support real-time, context-aware decision-making. The system leverages the LLM's reasoning and control capabilities to orchestrate device behaviors via standardized API-based interactions. The LGL-IoT networking framework is evaluated in a real-world testbed using emerging lightweight LLMs (e.g., DeepSeek-r1 and minimind2), demonstrating its capability in key metrics including sensor data fidelity, network latency, and end-to-end LLM responsiveness. Experimental results show that the LGL-IoT achieves efficient device coordination with reduced response time and resource overhead, offering a practical and scalable solution for LLM deployment in next-generation IoT environments.