c5.py

"""C5 dataset based on Common Crawl."""

import json
import io
import datasets
from tensorflow.io import gfile
import pyzstd


logger = datasets.logging.get_logger(__name__)


_DESCRIPTION = """\
A colossal, cleaned version of Common Crawl's web crawl corpus.
Based on Common Crawl dataset: "https://commoncrawl.org".
This is the processed version of Google's C5 dataset by AllenAI.
"""

_CITATION = """
@article{2019t5,
    author = {Colin Raffel and Noam Shazeer and Adam Roberts and Katherine Lee and Sharan Narang and Michael Matena and Yanqi Zhou and Wei Li and Peter J. Liu},
    title = {Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer},
    journal = {arXiv e-prints},
    year = {2019},
    archivePrefix = {arXiv},
    eprint = {1910.10683},
}
"""

_URL = "https://github.com/allenai/allennlp/discussions/5056"

_VARIANTS = ["en","other"]


buffer_size=1024
# _DATA_URL = "https://huggingface.co/datasets/allenai/c4/resolve/1ddc917116b730e1859edef32896ec5c16be51d0/{name}/c4-{split}.{index:05d}-of-{n_shards:05d}.json.gz"


def reverseMultiBytesIO(decompressedStream: io.BytesIO, buffer_size=1024):
    incomplete_line = bytearray()

    while True:
        # Read a chunk from the decompressed stream
        buffer = decompressedStream.read(buffer_size)  # Adjust the read size as needed
        if not buffer:
            # Edge Case 5: Handle EOF
            break

        # Concatenate with any incomplete line from the last read
        # Edge Case 1: Incomplete Lines
        buffer = incomplete_line + buffer
        incomplete_line = bytearray()

        # Split the buffer by the newline character
        lines = buffer.split(b'\n')
        # Check if the last line is complete
        if lines and lines[-1]:
            incomplete_line = lines.pop()

        # Decode and parse each line
        for line in lines:
            if line:
                yield json.loads(line.decode('utf-8'))
                

class C5Config(datasets.BuilderConfig):
    """BuilderConfig for C5."""

    def __init__(self, name, shard_id=None, subshard_id=None, buffer_size=1024, **kwargs):
        """BuilderConfig for C5.
        Args:
          **kwargs: keyword arguments forwarded to super.
        """
        super(C5Config, self).__init__(name=name,**kwargs)
        self.shard_id = shard_id
        self.subshard_id = subshard_id
        self.buffer_size = buffer_size


_DATA_URL_en = "gs://meliad2_us2/EasyLM/cluster/c5_{split}/shard_{shard_id}/subshard_{subshard_id}/text.jsonl"
_DATA_URL_non_en = "gs://meliad2_us2/EasyLM/v1/c5_{split}/shard_{shard_id}.jsonl"
class C5(datasets.GeneratorBasedBuilder):
    """C5, a colossal, cleaned version of Common Crawl's web crawl corpus."""

    BUILDER_CONFIGS = [C5Config(name) for name in _VARIANTS]

    def _info(self):
        self.shard_id = self.config.shard_id
        self.subshard_id = self.config.subshard_id
        self.buffer_size = self.config.buffer_size
        
        print(f"{self.name=}")
        if self.config.name=="en":
            features=datasets.Features(
                {
                    "text": datasets.Value("string"),
                    "timestamp": datasets.Value("string"),
                    "url": datasets.Value("string"),
                }
            )
        else:
            features=datasets.Features(
                {
                    "text": datasets.Value("string"),
                    "cluster_id": datasets.Value("int16"),
                }
            )
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
            supervised_keys=None,
            homepage=_URL,
            citation=_CITATION,
        )

    def _split_generators(self, dl_manager):
        data_urls = {}
        if self.config.name=="en":
            allowed_shards = range(64) if self.shard_id is None else [self.shard_id]
            allowed_subshards = range(64) if self.subshard_id is None else [self.subshard_id]
            all_allowed = [(shard_id, subshard_id) for shard_id in allowed_shards for subshard_id in allowed_subshards]
            format_data_url = lambda split,idx: _DATA_URL_en.format(split=split, shard_id=idx[0], subshard_id=idx[1])
        else:
            all_allowed = range(1024) if self.shard_id is None else [self.shard_id]
            format_data_url = lambda split,idx: _DATA_URL_non_en.format(split=split, shard_id=idx)
            
        for split in ["train"]:
            data_urls[split] = [
                format_data_url(split,idx) for idx in all_allowed
            ]
        train_downloaded_files = data_urls["train"]
        # train_downloaded_files = dl_manager.download()
        # validation_downloaded_files = dl_manager.download(data_urls["validation"])
        return [
            datasets.SplitGenerator(name=datasets.Split.TRAIN, gen_kwargs={"filepaths": train_downloaded_files}),
            # datasets.SplitGenerator(name=datasets.Split.VALIDATION, gen_kwargs={"filepaths": validation_downloaded_files}),
        ]

    def _generate_examples(self, filepaths):
        """This function returns the examples in the raw (text) form by iterating on all the files."""
        id_ = 0
        for filepath in filepaths:
            with gfile.GFile(filepath, 'rb') as f:
                logger.info("generating examples from = %s", filepath)
                with pyzstd.ZstdFile(f, 'rb') as ifo:
                    for example in reverseMultiBytesIO(ifo, buffer_size=self.buffer_size):
                        yield id_, example
                        id_ += 1