A benchmark for evaluating PDF parsing and document understanding systems.
[!WARNING]
Evaluation use only — do not train on this data.
This dataset is published as a held-out benchmark. Including it (or any
derivative of it) in model training or fine-tuning data contaminates the
benchmark and invalidates results. The data is provided in a single test
split for this reason; there is intentionally no train split.
If you operate a web crawler or assemble training corpora… See the full description on the dataset page:
https://huggingface.co/datasets/sleepyheeler/GDP.pdf.