license: odc-by
task_categories:
- text-generation
- text-classification
language:
- en
size_categories:
- 1M<n<10M
Dolma-1M
Dataset Description
An unofficial subset of Dolma v1.7 consisting of 1,000,000 training rows and about 100,000 testing rows, selected from 10 random files, each containing text entries between 500 and 5000 characters in length.
Advantages over the original Dolma dataset:
- much smaller and more manageable size
- filtered for length
- doesn't require remote executing Python scripts
Dataset Splitting
Training and testing sets were created by random sampling without replacement. The random sampling was done using Python. The dataset is stored in Gzipped JSONL format, where each row is a row from the original Dolma dataset.
Example line from the uncompressed file:
{
"id": "https://nightforvets.com/rex-lawrence-poutre/",
"text": "Rex was born on the family farm in Concordia, Kansas on February 6, 1920 to Arthur Donas Poutre and Ronalda Nadeau Poutre (Beland). Rex’s older brother, Leo, was born in 1917, and his younger brother, Bob, in 1925.\nAfter graduation from high school, Rex moved to Southern California, where his brother was stationed at March Field. [...] passion for riding motorcycles. He participated yearly, for 25 years, in the American Motorcycle Association’s two big tour-bike races, The Iron Butt and the Three Flags. He was a demon for speed. He will be missed.",
"added": "2023-04-10T09:48:38.760096+00:00",
"created": "2020-02-23T02:03:05Z",
"source": "common-crawl"
}
Dataset Usage
Feel free to use this dataset for research, experimentation, and development purposes. Like the original Dolma dataset, it's covered by the Open Data Commons Attribution License (ODC-By) v1.0.