DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation

Hugging Face Daily Papers Papers
visual-document-retrieval distillation efficient-ai retrieval asymmetric-encoder compact-model

Summary

DistilVDR is a compact 524M visual document retriever distilled from an 8B teacher via cosine alignment, achieving near-teacher accuracy on ViDoRe with 15.6x 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 smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector retriever end-to-end. We present DistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwise cosine alignment loss. All supervision comes from the frozen teacher's embedding space, which was itself trained with relevance supervision, so the student objective needs no relevance labels, negative sampling, or contrastive term. We match 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. We release two variants that share the same encoders and training and differ only in the document encoder's visual-tile budget: DistilVDR-HiRes attains 61.74 average NDCG@5 on ViDoRe v1+v2+v3 (86.9% of the 8B teacher) and leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark, while DistilVDR-Fast attains 59.98 at a 3 times smaller visual-token budget. Both variants store one million documents in a 15.6 times smaller index than the strongest sub-1B multi-vector baseline and index the corpus an order of magnitude faster. The code is available at https://github.com/Ryenhails/NanoVDR.
Original Article
View Cached Full Text

Cached at: 08/12/26, 08:19 AM

Paper page - DistilVDR: A Compact End-to-End Visual Document Retriever via Dual-Student Distillation

Source: https://huggingface.co/papers/2608.10636

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 smaller multi-vector encoder from scratch or distil only the query side; neither yields a compact single-vector retriever end-to-end. We presentDistilVDR, a 524M end-to-end VDR system distilled bilaterally from a single 8B vision-language teacher under a pointwisecosine alignment loss. All supervision comes from the frozen teacher’s embedding space, which was itself trained with relevance supervision, so the student objective needs no relevance labels, negative sampling, or contrastive term. We match VDR’s text-query and image-document input asymmetry with anasymmetric encoder-only student that concentrates visual capacity on the document side and keeps the query side at 70M parameters. We release two variants that share the same encoders and training and differ only in the document encoder’svisual-tile budget:DistilVDR-HiRes attains 61.74 averageNDCG@5onViDoRev1+v2+v3 (86.9% of the 8B teacher) and leads every reproduced sub-1B baseline on the high-resolution-sensitive v3 benchmark, whileDistilVDR-Fast attains 59.98 at a 3 times smaller visual-token budget. Both variants store one million documents in a 15.6 times smaller index than the strongest sub-1Bmulti-vector baselineand index the corpus an order of magnitude faster. The code is available at https://github.com/Ryenhails/NanoVDR.

View arXiv pageView PDFProject pageGitHub1Add to collection

Community

Paper submitter

about 3 hours ago

Upload images, audio, and videos by dragging in the text input, pasting, orclicking here.

Tap or paste here to upload images

Models citing this paper3

#### nanovdr/NanoVDR-D-Fast-Qwen3VL8B-4096 Feature Extraction• 0.5B• Updatedabout 3 hours ago • 2 #### nanovdr/NanoVDR-D-HiRes-Qwen3VL8B-4096 Feature Extraction• 0.5B• Updatedabout 3 hours ago • 1 #### nanovdr/NanoVDR-Q-DistilBERT-Qwen3VL8B-4096-ML Sentence Similarity• 66.4M• Updatedabout 3 hours ago • 1

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.10636 in a dataset README.md to link it from this page.

Spaces citing this paper1

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Visual Contrastive Self-Distillation

Hugging Face Daily Papers

VCSD removes the need for external teachers, privileged answers, or visual evidence in on-policy self-distillation by using content-erased control images to produce contrastive signals, consistently outperforming existing methods on vision-language benchmarks.