Skip to content

llm.preprocess.download

Pretraining corpus downloader.

python -m llm.preprocess.download --help

cli 

cli(
    dataset: Literal["wikipedia", "bookcorpus"],
    output_dir: str,
    log_level: str,
    rich: bool,
) -> None

Pretraining text downloader.

Source code in llm/preprocess/download.py 
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
@click.command()
@click.option(
    '-d',
    '--dataset',
    type=click.Choice(['wikipedia', 'bookscorpus'], case_sensitive=False),
    required=True,
    help='Dataset to download.',
)
@click.option(
    '-o',
    '--output-dir',
    metavar='PATH',
    required=True,
    help='Output directory.',
)
@click.option(
    '--log-level',
    default='INFO',
    type=click.Choice(
        ['DEBUG', 'INFO', 'WARNING', 'ERROR'],
        case_sensitive=False,
    ),
    help='Minimum logging level.',
)
@click.option(
    '--rich/--no-rich',
    default=False,
    help='Use rich output formatting.',
)
def cli(
    dataset: Literal['wikipedia', 'bookcorpus'],
    output_dir: str,
    log_level: str,
    rich: bool,
) -> None:
    """Pretraining text downloader."""
    init_logging(log_level, rich=rich)

    if dataset == 'wikipedia':
        download_wikipedia(output_dir)
    elif dataset == 'bookscorpus':
        download_bookscorpus(output_dir)
    else:
        raise AssertionError('Unreachable.')