DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation
- 类型:arxiv
- 标识:2608.10636
- 链接:https://arxiv.org/abs/2608.10636
- 主分类:rag
- 形态:method
- 被引:0
- 被引来源:Semantic Scholar
- S2被引:0
- 影响力被引:0
- TLDR:This work presents DistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwise cosine alignment loss and matches VDR's text-query and image-document input asymmetry with an asymmetric encoder-only student that concentrates visual capacity on the document side and keeps the query side at 70M parameters.
- 副分类:multimodal
- 待LLM分类:否
- 标题中文:DistilVDR:通过双学生蒸馏构建紧凑的端到端视觉文档检索器
- TLDR中文:本文提出 DistilVDR,一个 524M 的端到端 VDR 系统,通过逐点余弦对齐损失从一个 8B 视觉-语言教师模型进行双向蒸馏,并以非对称的纯编码器学生模型匹配 VDR 的文本查询与图像-文档输入不对称性,将视觉容量集中于文档端,查询端保持 70M 参数。
- 来源文件:
- /inbox/tom/_candidates/2026-08-12-agent-rag-longcontext-candidates.json
- /inbox/tom/_candidates/2026-08-13-agent-rag-longcontext-candidates.json
- [S2 enrich]