FLINT：查询与文档双侧固定预算的晚交互检索
原标题：FLINT: Fixed-budget Late Interaction
AI 摘要
研究团队提出 FLINT（Fixed-budget Late INTeraction）架构，用基于注意力的池化把查询和文档都压缩为固定数量的嵌入向量，从而解决 ColBERT 类多向量检索模型因逐 token 存储向量而导致的存储和延迟问题。在 BEIR 上，FLINT 效果与现有晚交互模型相当；在 BrowseComp+、CRUMB、TREC Tip-of-the-Tongue 2025 等长查询基准上优于 ConstBERT，端到端检索快 1.5–2.8 倍，索引相比 ColBERTv2 缩小 5.3–9.3 倍。
正文节选
FLINT: Fixed-budget Late Interaction Abstract Multi-vector retrieval models such as ColBERT often outperform single-vector models of comparable scale by matching queries and documents at the token level. However, retaining a vector for every token substantially increases storage requirements and slows retrieval compared to single-vector bi-encoders, limiting their scalability. We introduce FLINT (Fixed-budget Late INTeraction), an architecture that uses attention-based pooling to represent each