WARP加速多向量检索，效率提升41倍且保持高精度
WARP: An Efficient Engine for Multi-Vector Retrieval
- 动态相似度补全+隐式解压+两级聚合，三重优化检索效率
- 相比XTR参考实现降低41倍端到端延迟，比ColBERTv2/PLAID快3倍
- 适合需要高速精准检索的系统级应用，如大规模问答与信息检索
多向量检索方法如ColBERT及其变体ConteXtualized Token Retriever（XTR）虽具高精度，但在大规模场景下面临效率挑战。为此，我们提出WARP检索引擎，通过三项关键创新显著提升基于XTR目标训练的检索器效率：（1）WARP$_\text{SELECT}$实现动态相似度补全；（2）隐式解压，避免检索中耗时的向量重构；（3）两级压缩过程实现高效得分聚合。结合高度优化的C++内核，该系统相较XTR参考实现将端到端延迟降低41倍，相较于ColBERTv2/PLAID引擎提速3倍，同时保持原有检索质量。
原文摘要 · Abstract (English)
Multi-vector retrieval methods such as ColBERT and its recent variant, the ConteXtualized Token Retriever (XTR), offer high accuracy but face efficiency challenges at scale. To address this, we present WARP, a retrieval engine that substantially improves the efficiency of retrievers trained with the XTR objective through three key innovations: (1) WARP$_\text{SELECT}$ for dynamic similarity imputation; (2) implicit decompression, avoiding costly vector reconstruction during retrieval; and (3) a two-stage reduction process for efficient score aggregation. Combined with highly-optimized C++ kernels, our system reduces end-to-end latency compared to XTR's reference implementation by 41x, and achieves a 3x speedup over the ColBERTv2/PLAID engine, while preserving retrieval quality.
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。