BabyLM.py

import datasets

from typing import List

_DESCRIPTION = """\
Dataset for the shared baby language modeling task.
The goal is to train a language model from scratch on this data which represents
roughly the amount of text and speech data a young child observes.  
"""

_HOMEPAGE = "https://babylm.github.io"

filenames = [
    "aochildes.txt", 
    "bnc_spoken.txt",
    "cbt.txt",
    "children_stories.txt",
    "gutenberg.txt",
    "open_subtitles.txt",
    "qed.txt", 
    "simple_wikipedia.txt",
    "switchboard.txt",
    "wikipedia.txt"
]

class BabyLM(datasets.GeneratorBasedBuilder):
    
    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="original_strict_small",
            description="Small version of the dataset with 10M words, not cleaned",
            version="1.0.0",
        ),
        datasets.BuilderConfig(
            name="strict_small",
            description="Small version of the dataset with 10M words, cleaned",
            version="1.0.0",
        ),
        datasets.BuilderConfig(
            name="original_strict",
            description="Full version of the dataset with 10M words, not cleaned",
            version="1.0.0",
        ),
        datasets.BuilderConfig(
            name="strict",
            description="Full version of the dataset with 100M words, cleaned",
            version="1.0.0",
        ),
        datasets.BuilderConfig(
            name="strict_small_gold",
            description="Small version of the dataset with 10M words and gold POS tags",
            version="1.0.0",
        ),
        datasets.BuilderConfig(
            name="strict_gold",
            description="Full version of the dataset with 100M words and gold POS tags",
            version="1.0.0",
        )
    ]

    DEFAULT_CONFIG_NAME = "strict_small"

    def _info(self):
            features = datasets.Features(
                {
                    "text": datasets.Value("string"),
                    "tagged_text": datasets.Value("string"),
                    "filename": datasets.Value("string"),
                }
            )
            return datasets.DatasetInfo(
                # This is the description that will appear on the datasets page.
                description=_DESCRIPTION,
                features=features,  # Here we define them above because they are different between the two configurations
                homepage=_HOMEPAGE,
            )


    def _split_generators(self, dl_manager: datasets.DownloadManager) -> List[datasets.SplitGenerator]:
        """ 
        Returns data for different splits 
        """

        if "strict_small" in self.config.name:
            train_data_dir = "10M"
        else: 
            train_data_dir = "100M"
        if 'gold' in self.config.name:
            folder = 'tagged_gold'
        elif 'original' in self.config.name:
            folder = 'original'
        else:
            folder = 'tagged'

        urls_to_download = {
            "train": [f"{folder}/{train_data_dir}/{fn}" for fn in filenames],
            "dev": [f"{folder}/dev/{fn}" for fn in filenames],
            "test": [f"{folder}/test/{fn}" for fn in filenames]
        } 

        downloaded_files = dl_manager.download_and_extract(urls_to_download)

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "split": "train",
                    "filepaths": downloaded_files["train"]}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "split": "dev",
                    "filepaths": downloaded_files["dev"]}
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "split": "test",
                    "filepaths": downloaded_files["test"]
                }
            ),
        ]

     # method parameters are unpacked from `gen_kwargs` as given in `_split_generators`
    def _generate_examples(self, split, filepaths):
        # The `key` is for legacy reasons (tfds) and is not important in itself, but must be unique for each example.

        # the filepaths should be a list of filepaths 
        if isinstance(filepaths, str):
            filepaths = [filepaths]
        
        global_idx = 0 

        if 'original' in self.config.name:
            for filepath in filepaths:
                with open(filepath, encoding="utf-8") as f:
                    for line in f:
                        yield global_idx, {"text": line.strip(), "filename": '', 'tagged_text': line.strip()}
                        global_idx += 1

        else:
            for filepath in filepaths:
                with open(filepath, encoding="utf-8") as f:
                    is_tags = False
                    text = ""
                    filename = ""
                    # Every other row contains POS tags. First row is the filename (we can't use filepath since the file path changes upon caching)
                    for row in f:
                        if filename == "":
                            filename = row.strip()
                            continue
                        if is_tags:
                            yield global_idx, {"text": text.strip(), "tagged_text": row.strip(), "filename": filename}
                            global_idx += 1 
                            is_tags = False
                        else:
                            text = row
                            is_tags = True