Dataset · Hugging Face

ibm-research/VAREX

Name: VAREX
Creator: ibm-research
License: cdla-permissive-2.0
Keywords: task_categories:document-question-answering, task_categories:image-to-text, language:en, license:cdla-permissive-2.0, size_categories:1K<n<10K, format:parquet, format:optimized-parquet, modality:document, modality:image, modality:text, library:datasets, library:dask, library:polars, library:mlcroissant, arxiv:2603.15118

VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents VAREX (VARied-schema EXtraction) is a benchmark for evaluating multimodal foundation models on structured data extraction from government forms. It comprises 1,777 documents with 1,771 unique schemas across three structural categories, each provided in four input modalities. Ground truth is deterministic — generated via a Reverse Annotation pipeline that programmatically fills PDF templates with synthetic… See the full description on the dataset page: https://huggingface.co/datasets/ibm-research/VAREX.

View on Hugging FaceReport an issue

Author / organization	ibm-research
License (as declared)	cdla-permissive-2.0
Downloads	1.7K (as reported by Hugging Face)
Likes	7
Created	15 March 2026
Last updated at source	18 March 2026

document-extractionstructured-extractiondocument-aiform-understandingmultimodalbenchmarkjson-schema

Embed this entry

<iframe src="https://silytics.si/embed/item/be6d400d-1024-4557-b466-48a7e6b5cb49" width="420" height="190" style="border:0" loading="lazy" title="Silytics entry"></iframe>

Free, no tracking. More about feeds and embeds.

Popularity, last 2 days

1,712 now (+168 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.

Recent changes

16 Oct 2026 Description: VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents VAREX (VARied-schema EXtraction) is a benchmark for evaluating multimodal foundation models on structured data extraction from government forms. It comprises 1,777 docu → VAREX: A Benchmark for Multi-Modal Structured Extraction from Documents VAREX (VARied-schema EXtraction) is a benchmark for evaluating multimodal foundation models on structured data extraction from government forms. It comprises 1,777 docu

Changes detected by Silytics between refreshes of the source.

More from ibm-research

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 Information may be incomplete or out of date.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.