Dataset · Hugging Face

athrael-soju/vdr-train-500k

Name: vdr-train-500k
Creator: athrael-soju
License: mit
Keywords: task_categories:visual-document-retrieval, language:en, language:es, language:de, language:fr, language:it, language:pt, license:mit, size_categories:100K<n<1M, format:arrow, modality:image, modality:text, library:datasets, library:mlcroissant, region:us

VDR Train 500k VDR Train 500k pairs 499,704 search queries with the document-page images that answer them. Queries are in English (73%), Spanish, German, French, Italian and Portuguese. Construction The source collections are joined into one pool of query-page pairs. A vision encoder embeds every English page, PCA reduces the embeddings to 50 dimensions and K-Means splits them into 14 clusters. English pairs are sampled evenly across the clusters, each cluster… See the full description on the dataset page: https://huggingface.co/datasets/athrael-soju/vdr-train-500k.

View on Hugging FaceReport an issue

Author / organization	athrael-soju
License (as declared)	mit
Downloads	0 (as reported by Hugging Face)
Likes	0
Created	17 October 2026
Last updated at source	17 October 2026

document-retrievallate-interactioncolpalihard-negativesmultilingual

Embed this entry

<iframe src="https://silytics.si/embed/item/8d19b7c8-1ffc-4100-826c-b5722e45c83a" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Connected in the knowledge graph

athrael-soju/ColUltraVBERTModel · used to train
athrael-soju/BiUltraVBERTModel · used to train

Explore this entry in the graph

Recent changes

17 Oct 2026 Description: VDR Train 500k VDR Train 500k pairs 499,704 search queries with the document-page images that answer them, for training visual document retrievers. Queries are in English (73%), Spanish, German, French, Italian and Portuguese, and each come → VDR Train 500k VDR Train 500k pairs 499,704 search queries with the document-page images that answer them. Queries are in English (73%), Spanish, German, French, Italian and Portuguese. Construction The source collections are joined into on
17 Oct 2026 License: none → mit
17 Oct 2026 Description: none → VDR Train 500k VDR Train 500k pairs 499,704 search queries with the document-page images that answer them, for training visual document retrievers. Queries are in English (73%), Spanish, German, French, Italian and Portuguese, and each come
17 Oct 2026 Topics added task_categories:visual-document-retrieval, document-retrieval, late-interaction, colpali, hard-negatives, multilingual

Changes detected by Silytics between refreshes of the source.

Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 17 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.