number_reading.py

import os
import datasets

_CITATION = """\
# (Optional) Add your citation here
"""

_DESCRIPTION = """\
Number Reading
"""
LANGUAGES = [
    "english"
]
LEVELS = [
    "easy", "middle", "hard"
]
class NumberReadingConfig(datasets.BuilderConfig):
    def __init__(self, task_name, **kwargs):
        super().__init__(name=task_name, **kwargs)
        self.task_name = task_name


class NumberReading(datasets.GeneratorBasedBuilder):
    BUILDER_CONFIGS = [
        NumberReadingConfig(task_name=level) for level in LEVELS
    ]
    
    def _info(self):
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=datasets.Features({
                "golden_reading": datasets.Sequence(datasets.Value("string")),
                "prompt": datasets.Value("string"),
                "number": datasets.Value("string"),
                "language": datasets.Value("string"),
                # add more fields depending on your JSONL schema
            }),
            supervised_keys=None,
            homepage="https://huggingface.co/datasets/huayangli/number_reading",
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        level = self.config.name
        data_files = {
            "test": dl_manager.download_and_extract(f"english_{level}_testdata_manyshot_prompt.jsonl"),
        }
        return [
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={"filepath": data_files["test"]},
            )
        ]
    def _generate_examples(self, filepath):
        import json
        with open(filepath, "r", encoding="utf-8") as f:
            for idx, line in enumerate(f):
                data = json.loads(line)
                data.pop("training")
                yield idx, data