AWhat changed
FineWeb documented and ablated filtering/deduplication for a large English web corpus and an educational subset; FineWeb2 adapted curation statistics by language and rebalanced globally deduplicated multilingual data using duplication and quality signals.
Scientific scope
FineWeb documented and ablated filtering/deduplication for a large English web corpus and an educational subset; FineWeb2 adapted curation statistics by language and rebalanced globally deduplicated multilingual data using duplication and quality signals. One source-specific family, not separate invention credit for FineWeb-Edu or each language. Inherits Common Crawl and earlier filters/classifiers. FineWeb2's broad language coverage is not evidence of uniformly high quality or equal gains in every language; ablations cover a limited language/model suite. Dataset sizes, tokenization and benchmark gains are version-specific; no content, rights or contamination clearance inferred.
Evidence
The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. One source-specific family, not separate invention credit for FineWeb-Edu or each language. Inherits Common Crawl and earlier filters/classifiers. FineWeb2's broad language coverage is not evidence of uniformly high quality or equal gains in every language; ablations cover a limited language/model suite. Dataset sizes, tokenization and benchmark gains are version-specific; no content, rights or contamination clearance inferred.
Contributors
- Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.
- The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale
- FineWeb2: One Pipeline to Scale Them All -- Adapting Pre-Training Data Processing to Every Language
Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.
Full scientific record · Browse contribution grades · Request a correction