为分析型数据库定制轻量LLM，大幅降低计算成本。
The Case for Instance-Optimized LLMs in OLAP Databases
- 针对每个查询生成专用小模型，用样本数据优化
- 模型体积减少76%，吞吐量提升3.31倍，精度不变
- 适合需要大规模语言模型的数据库分析场景
大型语言模型（LLMs）可为分析系统提供强大的数据摘要、清洗和语义转换能力。然而，在数百万至数十亿行数据上规模化部署LLM仍面临巨大的计算与内存开销。本文提出IOLM-DB，一种通过查询特异性模型优化使LLM增强型数据库查询可行的新系统。不同于通用大模型，IOLM-DB利用代表性数据样本为每个查询生成轻量、专精的模型。该方法结合量化、稀疏化和结构化剪枝等激进压缩技术，将模型规模减少高达76%，吞吐量提升最高达3.31倍，同时保持准确率。我们还展示了该方法如何在现有硬件上实现更高并行度，并无缝支持缓存与批处理策略以降低开销。原型系统验证了在分析系统中大规模使用LLM查询的可行性，为未来OLAP应用开辟新可能。
原文摘要 · Abstract (English)
Large Language Models (LLMs) can enhance analytics systems with powerful data summarization, cleaning, and semantic transformation capabilities. However, deploying LLMs at scale -- processing millions to billions of rows -- remains prohibitively expensive in computation and memory. We present IOLM-DB, a novel system that makes LLM-enhanced database queries practical through query-specific model optimization. Instead of using general-purpose LLMs, IOLM-DB generates lightweight, specialized models tailored to each query's specific needs using representative data samples. IOLM-DB reduces model footprints by up to 76% and increases throughput by up to 3.31$\times$ while maintaining accuracy through aggressive compression techniques, including quantization, sparsification, and structural pruning. We further show how our approach enables higher parallelism on existing hardware and seamlessly supports caching and batching strategies to reduce overheads. Our prototype demonstrates that leveraging LLM queries inside analytics systems is feasible at scale, opening new possibilities for future OLAP applications.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。