UIBK-Embedding：统一稠密与延迟交互检索
原标题：UIBK-Embedding: Effective Text Retrieval by Unifying Dense and Late-Interaction Retrieval
AI 摘要
研究团队提出 UIBK-Embedding-v1，一个 179M 参数的英文编码器，在一次前向中同时生成稠密向量和 token 级向量，并用未修改的 MaxSim 统一打分，从而在标准 late-interaction 索引（PLAID）中同时获得两种检索能力。在 BEIR 上达到 59.79 nDCG@10，比同规模最强模型 LateOn 高 2.57 分，并在全部 15 个数据集上领先；在去污染 BEIR 上仍以 61.55 对 60.36 保持优势。在 MTEB 结果库中，只有参数量至少 1.5B 的模型得分更高。代码、权重和技术博客已开源。
正文节选
UIBK-Embedding: Effective Text Retrieval by Unifying Dense and Late-Interaction Retrieval Abstract Dense bi-encoders and late-interaction retrievers are usually treated as alternatives, yet they fail on different queries: one vector captures what a text is about, token-level matching finds its exact entities and phrases. Using both today means two models, two indexes and a score-fusion step. We present UIBK-Embedding-v1, a 179M-parameter English encoder that produces both representations in one