LEGAR_BENCH.py

import json
import datasets

_DESCRIPTION = """\
LEGAR_BENCH is the first large-scale Korean LCR benchmark, covering 411 diverse crime types in queries over 1.2M legal cases.
"""

_HOMEPAGE = "https://huggingface.co/datasets/Chaeeun-Kim/LEGAR_BENCH"
_LICENSE = "Apache 2.0"

_URLS = {
    "standard": "data/standard_train.jsonl",
    "stricter": "data/stricter_train.jsonl", 
    "stricter_by_difficulty": "data/stricter_by_difficulty_train.jsonl",
}

class LegarBench(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.0.0")
    
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="standard",
            version=VERSION,
            description="Standard version of LEGAR BENCH",
        ),
        datasets.BuilderConfig(
            name="stricter",
            version=VERSION, 
            description="Stricter version of LEGAR BENCH",
        ),
        datasets.BuilderConfig(
            name="stricter_by_difficulty",
            version=VERSION,
            description="Stricter version organized by difficulty",
        ),
    ]

    DEFAULT_CONFIG_NAME = "standard"

    def _info(self):
        features = datasets.Features({
            "id": datasets.Value("int64"),
            "target_category": datasets.Value("string"),
            "category": datasets.Value("string"),
            "question": datasets.Value("string"),
            "question_id": datasets.Value("string"),
            "answer": datasets.Sequence(datasets.Value("string")),
            "evidence_id": datasets.Sequence(datasets.Value("string")),
            "difficulty": datasets.Value("string"),
        })

        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            license=_LICENSE,
        )

    def _split_generators(self, dl_manager):
        url = _URLS[self.config.name]
        data_file = dl_manager.download(url)
        
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "filepath": data_file,
                },
            ),
        ]

    def _generate_examples(self, filepath):
        with open(filepath, encoding="utf-8") as f:
            for key, line in enumerate(f):
                data = json.loads(line)
                
                yield key, {
                    "id": int(data.get("id", 0)),
                    "target_category": str(data.get("target_category", "")),
                    "category": self._process_category(data.get("category", {})),
                    "question": str(data.get("question", "")),
                    "question_id": str(data.get("question_id", "")),
                    "answer": self._process_list_field(data.get("answer", [])),
                    "evidence_id": self._process_list_field(data.get("evidence_id", [])),
                    "difficulty": str(data.get("difficulty", "")),
                }
    
    def _process_category(self, category):
        if isinstance(category, dict):
            return json.dumps(category, ensure_ascii=False)
        return str(category)
    
    def _process_list_field(self, field):
        if field is None:
            return []
        return [str(item) if item is not None else "" for item in field]