microsoft/XL-DocBench
XL-DocBench Evidence-grounded reasoning across hundreds or thousands of pages. Fully verified by 194 human experts. Hongchen Wei1,†,‡, Yuanzhe Wang2,†,‡, Bei Liu2,*, Yifan Yang2, Qi Dai2, Ruichun Ma2, Kai Qiu2, Yunsheng Li2, Dongdong Chen2, Chong Luo2, Zhenzhong Chen1, Baining Guo2 1Wuhan University 2Microsoft †Equal contribution ‡Work done during an internship at MSRA *Project leader Project Page · Paper · Live Leaderboard… See the full description on the dataset page: https://huggingface.co/datasets/microsoft/XL-DocBench.
7944
1#!/usr/bin/env python32"""Load XL-DocBench JSONL files with the Python standard library."""3 4from __future__ import annotations5 6import argparse7import json8from pathlib import Path9 10 11def parse_args() -> argparse.Namespace:12 parser = argparse.ArgumentParser(description=__doc__)13 parser.add_argument(14 "--release-root",15 type=Path,16 default=Path(__file__).resolve().parents[2],17 help="Release directory; defaults to the directory containing code/",18 )19 return parser.parse_args()20 21 22def read_jsonl(path: Path) -> list[dict]:23 with path.open("r", encoding="utf-8") as handle:24 return [json.loads(line) for line in handle if line.strip()]25 26 27def main() -> None:28 root = parse_args().release_root.resolve()29 documents = read_jsonl(root / "data/documents.jsonl")30 single_doc = read_jsonl(root / "data/qa_single_doc.jsonl")31 cross_doc = read_jsonl(root / "data/qa_cross_doc.jsonl")32 33 print(f"documents: {len(documents):,}")34 print(f"single_doc: {len(single_doc):,}")35 print(f"cross_doc: {len(cross_doc):,}")36 print(f"first question: {single_doc[0]['question']}")37 38 39if __name__ == "__main__":40 main()41