MedDocBench Sample: 10 Clinical Documents for Vision-Language Model Testing

MedDocBench Sample is a public 10-document release for benchmarking vision-language models on structured extraction from real-world clinical forms. The sample is provided by Owenhku for repository smoke tests, format inspection, and reproducibility checks. The full secondary-redacted dataset is released as a separate gated Hugging Face dataset.

Use Cases

• Benchmarking structured extraction performance based on real-world clinical forms mentioned in the description
• Conducting format inspection and reproducibility checks for data pipelines based on the sample's purpose
• Testing vision-language model reliability on medical documents based on the dataset's stated objective

Strengths

• Publicly available 10-document sample for initial inspection
• Specifically designed for benchmarking vision-language models on a real-world task

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment
• Column-level documentation is absent; field semantics must be inferred after download
• Description metadata is limited; actual data quality requires manual inspection after download

Provenance

Source

Owenhku on Hugging Face

Collection Method

Likely derived from real-world clinical forms for benchmarking purposes.

Freshness

Last updated 2026-06-20 14:41:53; freshness should be verified