The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets
library). If this is not possible, please
open a discussion
for direct help.
This is a reproduction of the CC-stories dataset as it has been removed from its original source. To create this reproduction we process the English common crawl and only keep the top 0.1% of documents measured by their ngram overlap with a source document. The source document is created by joining the queries from PDP-60 and WSC273. Note, as the original dataset does not mention removing duplicate queries, neither do we.
Following the filtering to have top documents we filter to only contain those and produce the dataset which features 2,105,303 lines and 153,176,685 words.
- Downloads last month
- 544