This dataset, HR, is a corpus of reports released by the european union, intended for complex-document understanding tasks. It is one of the 10 corpora comprising the ViDoRe v3 Benchmark.
ViDoRe V3 is our latest benchmark for RAG evaluation on visually-rich documents from real-world applications. It features 10 datasets with, in total, 26,000 pages and 3099 queries, translated into 6 languages. Each query comes with human-verified relevant pages… See the full description on the dataset page:
https://huggingface.co/datasets/vidore/vidore_v3_hr.
