NetGPT：超越个性化生成服务的内生智能网络架构

摘要：提出了基于边缘和云端部署相匹配大型语言模型（LLM）的内生智能网络架构（NetGPT）方案。边缘LLM可以有效地利用基于位置的信息进行个性化的补充，从而与云端LLM进行有效交互。通过在边缘和云端部署开源LLM，验证了NetGPT的可行性。认为面向NetGPT的内生智能网络架构的工作重点是通信和计算资源的深度集成以及AI逻辑工作流的灵活设计。认为NetGPT是一种可提供个性化的生成式服务的、有前途的内生智能网络架构。

关键词：LLM；内生智能网络架构；云边协同；个性化生成服务

Abstract: The NetGPT framework, which is founded upon the alignment of large language models (LLMs) tailored for both edge and cloud deployments is introduced. Edge-oriented LLMs harness location-based data to effectively personalize content augmentation, facilitating seamless interactions with their cloud-based counterparts. The viability of the NetGPT paradigm is empirically substantiated through the deployment of open-source LLMs at both the edge and cloud strata. It is believed that within the realm of endogenous intelligent network architectures designed to support NetGPT, the central emphasis rests on the profound integration of communication and computational resources, coupled with the adaptability in the design of AI logic workflows.

Keywords: LLM; AI-native network architecture; edge-cloud collaboration; personalized generative services