proof-pile-2.py

import json
import os
import zstandard as zstd

import datasets

_CITATION="""\
@article{azerbayev2023llemma,
    title={Llemma: an open language model for mathematics},
    author={Zhangir Azerbayev and Hailey Schoelkopf and Keiran Paster and Marco Dos Santos and Stephen McAleer and Albert Q. Jiang and Jia Deng and Stella Biderman and Sean Welleck},
    eprint={xyz.xyz},
    archivePrefix={arXiv}
    year={2023}
}
"""

_DESCRIPTION = """\
A dataset of high quality mathematical text. """
_HOMEPAGE = "https://github.com/EleutherAI/math-lm"


SPLITS = ["train", "validation", "test"]

_DATA_PATHS = {
        "arxiv": {
            split: [f'arxiv/{split}/arXiv_{str(i).zfill(3)}.jsonl.zst' for i in range(100)]
            for split in SPLITS        
        },
        "open-web-math": {
            split: [
                os.path.join(f"open-web-math/{split}", filename)
                for filename in os.listdir(f"open-web-math/{split}")
            ] 
            for split in SPLITS
        },
        "algebraic-stack": {
            split: [
                os.path.join(f"algebraic-stack/{split}", filename)
                for filename in os.listdir(f"algebraic-stack/{split}")
            ] 
            for split in SPLITS
        }
}

class ProofPile2Config(datasets.BuilderConfig):
    """BuilderConfig for RedPajama sample."""

    def __init__(self, *args, subsets, **kwargs):
        """BuilderConfig for ProofPile2.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(ProofPile2Config, self).__init__(**kwargs)
        self.subsets = subsets


class ProofPile2(datasets.GeneratorBasedBuilder):
    """A large dataset of mathematical text."""
    VERSION = datasets.Version("1.1.0")

    # This is an example of a dataset with multiple configurations.
    # If you don't want/need to define several sub-sets in your dataset,
    # just remove the BUILDER_CONFIG_CLASS and the BUILDER_CONFIGS attributes.

    # If you need to make complex sub-parts in the datasets with configurable options
    # You can create your own builder configuration class to store attribute, inheriting from ProofPile2Config
    # BUILDER_CONFIG_CLASS = MyBuilderConfig

    # You will be able to load one or the other configurations in the following list with
    # data = datasets.load_dataset('my_dataset', 'first_domain')
    # data = datasets.load_dataset('my_dataset', 'second_domain')
    BUILDER_CONFIGS = [
        ProofPile2Config(
            name='default',
            subsets=list(_DATA_PATHS.keys()),
            version=VERSION,
            description="All subsets"
        ),
        ProofPile2Config(
            name='arxiv',
            subsets=["arxiv"],
            version=VERSION,
            description="ArXiv subset"
        ),
        ProofPile2Config(
            name='open-web-math',
            subsets=['open-web-math'],
            version=VERSION,
            description="OpenWebMath"
        ),
        ProofPile2Config(
            name='algebraic-stack',
            subsets=['algebraic-stack'],
            version=VERSION,
            description="Code subset"
        ),
    ]


    def _info(self):
        features = datasets.Features(
            {
                "text": datasets.Value("string"),
                "meta": datasets.Value("string")
            }
        )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            homepage=_HOMEPAGE,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        return [ 
            datasets.SplitGenerator(
                name=split_obj,
                # These kwargs will be passed to _generate_examples
                gen_kwargs={
                    "data_files": list(map(
                        dl_manager.download, 
                        [
                            x 
                            for subset in self.config.subsets 
                            for x in dl_manager.iter_files(os.path.join(subset, split))
                        ]
                    ))
                },
            )
            for split, split_obj in zip(
                ("train", "validation", "test"),
                (datasets.Split.TRAIN, datasets.Split.VALIDATION, datasets.Split.TEST)
            )
        ]

    def _generate_examples(self, data_files):
        key = 0 
        for name in data_files: 
            with zstd.open(open(name, "rb"), "rt", encoding="utf-8") as f: 
                for x in f.readlines():
                    instance = json.loads(x)
                    if instance:
                        if "meta" not in instance:
                            instance["meta"] = dict()
                        yield key, {"text": instance["text"], "meta": json.dumps(instance["meta"])}
                        key += 1