1、 研究报告 （2024 年第 18 期 总第 50 期）2024 年 12 月 18 日 LLMLLM 时代小模型的应用潜力与挑战时代小模型的应用潜力与挑战 清华大学五道口金融学院财富管理研究中心 张晓燕 张艺伟 张远远 【摘要摘要】大语言模型在自然语言处理领域取得了显著成就。然而，由于其高昂的成本和资源消耗，其商业化应用仍处于早期阶段。相比之下，小语言模型凭借较低的训练成本和较小的计算资源需求等优势，展现出很大的潜力。本研究通过分析国内外小模型的发展现状与具体案例探讨了小模型的应用潜力与面临的挑战。研究发现：（1）小模型体型虽小但功能强大，具有成本优势；（2）我国小模型在专业领域中的表现能够

2、超越国外通用模型，同时综合能力大幅提升，具有技术创新性；（3）由于大模型对算 力资源的高需求，我国在资源与技术受限的情况下发展小模型，能够助力我国在全球 AI 竞赛中脱颖而出。本文为我国 AI 行业发展提供了战略方向。Research Report December 18,2024 Application Potential and Challenges of Small Language Model in the LLM Era Research Center for Wealth Management,PBC School of Finance,Tsinghua University Zh

3、ang Xiaoyan Zhang Yiwei Zhang Yuanyuan Abstract:Large language models have made remarkable achievements in the field of natural language processing.However,due to their high cost and resource consumption,their commercial application is still in its early stages.In contrast,small language models have

4、 shown great potential due to their advantages such as low training cost and small computing resource requirements.This study explores the application potential and challenges of small models by analyzing the development status and specific cases of small models at home and abroad.The study found th

5、at:(1)Small models are small in size but powerful in function,and have cost advantages;(2)my countrys small models can outperform foreign general models in professional fields,and their comprehensive capabilities are greatly improved,which is technologically innovative;(3)Due to the high demand for

6、computing resources by large models,my countrys development of small models under the condition of limited resources and technology can help my country stand out in the global AI competition.This article provides a strategic direction for the development of my countrys AI industry.目录目录 一、小模型的发展背景及意义