摘要
近年来大模型技术方兴未艾,在通用领域获得长足发展。然而,在军事、政务等关键领域训练数据不足导致专业领域的大模型应用能力难以满足用户的需求,特别是智能代理功能明显衰弱,难以理解业务需求并优选相应模型,导致在业务领域大模型调度能力存在瓶颈。本文提出一种面向业务的大模型智能代理框架,通过业务模板配置,捕捉用户需求,映射业务流程,智能调度业务资源(智能模型、系统功能、数据、知识等),有效缓解了用户意图难理解、通用模型难以快速支持复杂任务的问题。设计原型系统,在典型场景下进行验证,试验结果表明,面向业务微调的方法能够有效提升大模型在复杂业务流程调度的精准性。
In recent years,large language model(LLM)technology has been flourishing and has made significant progress in general fields.However,the lack of training data in key areas such as military and government affairs has made it difficult for large models in specialized domains to meet user needs,particularly in terms of weakened intelligent agent functions that struggle to understand business requirements and select appropriate models,leading to a bottleneck in the dispatching capability of large models in business areas.This paper proposes a business-oriented intelligent agent framework for large language models,which,through business template configuration,captures user needs,maps business processes,and intelligently dispatches business resources(intelligent models,system functions,data,knowledge,etc.),effectively alleviating the issues of difficult user intent understanding and the challenge of general models rapidly supporting complex tasks.A prototype system is designed and validated in typical scenarios.Experimental results show that the business-oriented fine-tuning approach can effectively enhance the precision of large models in dispatching complex business processes.
作者
张永
纪威宇
姜巍
ZHANG Yong;JI Weiyu;JIANG Wei(Information Science Academy,China Electronics Technology Group Corporation,Beijing 100042)
出处
《软件》
2024年第6期100-102,共3页
Software
关键词
大模型
智能代理框架
智能调度
large language model
intelligent agent framework
intelligent dispatching
作者简介
张永(1992-),男,河北邯郸人,硕士,工程师,研究方向:大模型、知识图谱。
