Drift-Adapter：在向量数据库中实现近零停机嵌入模型升级的实用方法
Drift-Adapter: A Practical Approach to Near Zero-Downtime Embedding Model Upgrades in Vector Databases
AI总结：
本文提出轻量可学习转换层 Drift-Adapter，将新查询映射到旧嵌入空间以复用 ANN 索引，实验表明其以极低延迟和成本恢复 95%–99% 的检索效果。
摘要：
在生产环境向量数据库中升级嵌入模型通常需要重新编码整个语料库并重建近似最近邻（ANN）索引，从而造成显著的运营中断和计算成本。本文提出 Drift-Adapter，这是一个轻量级、可学习的转换层，旨在连接不同模型版本之间的嵌入空间。通过将新查询映射到旧版嵌入空间，Drift-Adapter 使现有 ANN 索引能够继续使用，从而有效推迟全面重新计算。我们系统评估了三种适配器参数化形式：Orthogonal Procrustes、Low-Rank Affine 和紧凑的 Residual MLP，它们在少量成对的新旧嵌入样本上进行训练。在 MTEB 文本语料库和一次 CLIP 图像模型升级（100 万个项目）上的实验表明，Drift-Adapter 可恢复完全重新嵌入所达到检索召回率（Recall@10、MRR）的 95%–99%，并且增加的查询延迟低于 10 微秒。与完全重建索引或双索引服务等运营策略相比，Drift-Adapter 将重计算成本降低超过 100 倍，并支持以近零运营中断完成升级。我们分析了其对不同模型漂移、训练数据规模的鲁棒性、扩展至十亿级项目系统的能力，以及对角缩放等设计选择的影响，证明 Drift-Adapter 作为敏捷模型部署的务实解决方案是可行的。
英文摘要：
Upgrading embedding models in production vector databases typically requires re-encoding the entire corpus and rebuilding the Approximate Nearest Neighbor (ANN) index, leading to significant operational disruption and computational cost. This paper presents Drift-Adapter, a lightweight, learnable transformation layer designed to bridge embedding spaces between model versions. By mapping new queries into the legacy embedding space, Drift-Adapter enables the continued use of the existing ANN index, effectively deferring full re-computation. We systematically evaluate three adapter parameterizations: Orthogonal Procrustes, Low-Rank Affine, and a compact Residual MLP, trained on a small sample of paired old and new embeddings. Experiments on MTEB text corpora and a CLIP image model upgrade (1M items) show that Drift-Adapter recovers 95-99% of the retrieval recall (Recall@10, MRR) of a full re-embedding, adding less than 10 microseconds of query latency. Compared to operational strategies like full re-indexing or dual-index serving, Drift-Adapter reduces recompute costs by over 100 times and facilitates upgrades with near-zero operational interruption. We analyze robustness to varied model drift, training data size, scalability to billion-item systems, and the impact of design choices like diagonal scaling, demonstrating Drift-Adapter's viability as a pragmatic solution for agile model deployment.
机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。
