A document-extraction QA benchmark across 4 domains: finance, medical/healthcare,
mortgage, supply chain. Each item is a single-document question with a typed
JSON gold answer.
1359 questions across 581 documents
Domains: finance, medical_healthcare, mortgage, supply_chain
Bank shape: each item has question, typed template, gold_dict,
capabilities (137 tags grouped into 8 capability buckets), source_file,
domain.
Layout… See the full description on the dataset page: https://huggingface.co/datasets/Extend-AI/RealDoc-Bench.
