📄 中文摘要
该研究提出了一种新的缓存架构，旨在通过验证感知的多层缓存机制，显著降低大型语言模型（LLM）的运行成本和延迟。通过实施这一架构，研究显示可以将LLM的成本降低30%。该方法不仅提高了系统的响应速度，还优化了资源的使用效率，适应了大规模应用的需求。研究结果表明，采用这种零浪费的智能RAG策略能够有效提升LLM的性能，同时减少不必要的开支，为未来的AI应用提供了可行的解决方案。
零浪费智能RAG：设计缓存架构以在规模上最小化延迟和LLM成本
📄 English Summary
Zero-Waste Agentic RAG: Designing Caching Architectures to Minimize Latency and LLM Costs at Scale
A novel caching architecture is proposed to significantly reduce the operational costs and latency of large language models (LLMs) through validation-aware, multi-tier caching mechanisms. The implementation of this architecture demonstrates a potential cost reduction of 30% for LLMs. This approach not only enhances system response times but also optimizes resource utilization, catering to the demands of large-scale applications. The findings indicate that adopting this zero-waste agentic RAG strategy can effectively improve LLM performance while minimizing unnecessary expenditures, providing a viable solution for future AI applications.