leideng/longbench-v2-view
LongBench v2: Towards Deeper Understanding and Reasoning on Realistic Long-context Multitasks 🌐 Project Page: https://longbench2.github.io 💻 Github Repo: https://github.com/THUDM/LongBench 📚 Arxiv Paper: https://arxiv.org/abs/2412.15204 LongBench v2 is designed to assess the ability of LLMs to handle long-context problems requiring deep understanding and reasoning across real-world multitasks. LongBench v2 has the following features: (1) Length: Context length ranging from 8k… See the full description on the dataset page: https://huggingface.co/datasets/leideng/longbench-v2-view.
058
1#!/usr/bin/env python32 3import argparse4import json5import re6import unicodedata7from collections import defaultdict8from pathlib import Path9 10 11def slugify(value: str) -> str:12 normalized = unicodedata.normalize("NFKD", value)13 ascii_value = normalized.encode("ascii", "ignore").decode("ascii")14 slug = re.sub(r"[^a-zA-Z0-9]+", "-", ascii_value.lower()).strip("-")15 return slug or "unknown"16 17 18def load_records(input_path: Path) -> list[dict]:19 with input_path.open("r", encoding="utf-8") as f:20 data = json.load(f)21 22 if not isinstance(data, list):23 raise ValueError(f"{input_path} must contain a top-level JSON array.")24 25 for index, item in enumerate(data):26 if not isinstance(item, dict):27 raise ValueError(f"Record at index {index} is not a JSON object.")28 29 return data30 31 32def group_records(records: list[dict]) -> dict[tuple[str, str], list[dict]]:33 grouped: dict[tuple[str, str], list[dict]] = defaultdict(list)34 35 for index, record in enumerate(records):36 domain = record.get("domain")37 sub_domain = record.get("sub_domain")38 39 if not domain or not sub_domain:40 raise ValueError(41 f"Record at index {index} is missing a non-empty 'domain' or 'sub_domain'."42 )43 44 grouped[(str(domain), str(sub_domain))].append(record)45 46 return grouped47 48 49def write_groups(grouped_records: dict[tuple[str, str], list[dict]], output_dir: Path) -> None:50 for (domain, sub_domain), records in grouped_records.items():51 domain_dir = output_dir / slugify(domain)52 output_path = domain_dir / f"{slugify(sub_domain)}.json"53 54 domain_dir.mkdir(parents=True, exist_ok=True)55 with output_path.open("w", encoding="utf-8") as f:56 json.dump(records, f, ensure_ascii=False, indent=2)57 f.write("\n")58 59 60def parse_args() -> argparse.Namespace:61 parser = argparse.ArgumentParser(62 description="Split a JSON array into data/<domain>/<sub-domain>.json files."63 )64 parser.add_argument(65 "--input",66 type=Path,67 default=Path("data.json"),68 help="Path to the source JSON file. Default: data.json",69 )70 parser.add_argument(71 "--output-dir",72 type=Path,73 default=Path("data"),74 help="Directory where split files will be written. Default: data",75 )76 return parser.parse_args()77 78 79def main() -> None:80 args = parse_args()81 records = load_records(args.input)82 grouped_records = group_records(records)83 write_groups(grouped_records, args.output_dir)84 85 86if __name__ == "__main__":87 main()88