license: odc-by
Description of Dataset
This is a subset of the Falcon Refinedweb dataset, with tagging performed using the taggers in the ontocord/riverbed repository. This subset includes approximately 100K items (485MB or approximately 100M tokens). The items have been filtered for moderately high Red-Pajama 1 filter scores (0.65) and Textbook quality scores (0.65). Tags are added to the beginning of some of the items when they have been detected to be science based, how-to, legal, etc. The original tagging was performed by Ontocord for the Aurora-M project, and subsequently further tagged for textbook quality by @kenhktsui here. It was further filtered to create this subset.
Original Refinedweb Dataset
The authors of Refinedweb writes:
RefinedWeb is built through stringent filtering and large-scale deduplication of CommonCrawl; we found models trained on RefinedWeb to achieve performance in-line or better than models trained on curated datasets, while only relying on web data. See the 📓 paper on arXiv for more details on Refinedweb itself.
Purpose
The purpose of the dataset is to investigate filtering of data and creating performant mixtures of data for model training.
Disclaimer
This dataset is provided for research purposes only. We make no claims to ownership of the data and release any of our annotations under ODC-By because that was the original RefinedWeb license. We believe usage of web crawled data is permitted under fair use, but you use this dataset at your own risks, and we disclaim all warranties and liabilities including warranties for non-infringement.
NO PERSONALLY IDENTIFIABLE INFORMATION REDACTION WAS PERFORMED. ALL SUCH INFORMATION ARE FROM THE ORIGINAL DATASET AND THE CORRESPONDING WEB PAGE.
Contact Us
Please reach out to us in the discussion link above if you have questions.