FlexVDR：为视觉文档检索学习分配套娃表示预算
原标题：FlexVDR: Learning to Allocate Matryoshka Representation Budgets for Visual Document Retrieval
AI 摘要
论文提出 FlexVDR 框架，在单一检索器内为每个查询自适应分配表示预算。它构建了联合覆盖编码器深度、嵌入宽度和 token 数量的路由族，并用 Budget Allocator 根据早期查询特征预测各路由的检索效用，再结合可配置的资源偏好选择路由。在 ViDoRe V1/V2/V3 上跨五个骨干模型的实验显示，该方法在多数情况下保留原生检索质量，同时可为单个查询选择紧凑路由，且无需重训练或切换模型即可调整计算量。
正文节选
FlexVDR: Learning to Allocate Matryoshka Representation Budgets for Visual Document Retrieval Abstract Visual document retrieval relies on rich multi-vector representations to capture fine-grained evidence in document pages, but encoding and matching these representations can be computationally expensive. Existing compression and elastic representation methods expose smaller configurations, yet making multiple budgets available does not determine which configuration a query should use. The allo