DistilVDR：通过双学生蒸馏实现紧凑的端到端视觉文档检索器

AI 摘要

DistilVDR 是一个紧凑的 524M 视觉文档检索器，通过双学生蒸馏从 8B 教师模型中提炼而来，使用余弦对齐损失，无需相关性标签。它在 ViDoRe 基准上达到教师模型 86.9% 的 NDCG@5 性能，同时索引体积缩小 15.6 倍，索引速度提升一个数量级。研究团队发布了训练代码和模型权重。

正文节选

DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation Abstract DistilVDR is a compact 524M vision-document retriever distilled from an 8B teacher using cosine alignment without relevance labels, achieving near-teacher accuracy with far smaller indexes and faster indexing. Visual document retrieval (VDR) is dominated by multi-billion-parameter models that are slow to index at full corpus scale and expensive to serve. Prior compression routes either train a smalle

发布时间：—

抓取时间：2026-08-12 14:22

来源机构：Hugging Face