Updated training/validation set
[17/02]: The training and validation dataset has been updated to also contain list and unanswerable questions. Question-answer pairs from the previous subset have been revised.
The links under Download have been updated to point to the newest annotations file, as well as to the binaries with some patched OCR files.
For the easiest access to the latest version of the dataset, we suggest you use DUDE_loader on HuggingFace Datasets.