FineWeb2-Edu-Scores: Multilingual Education Quality Scores for Language Model Training

FineWeb2-JQL-Education is a model-annotated language subset of the FineWeb2 dataset, spanning 36 languages. The dataset was created by JQL-AI and last updated on August 7, 2025. It provides scores intended to filter web data for higher-quality training outcomes without excessively aggressive data reduction.

Use Cases

• Filtering web-crawled text for model training based on model-annotated education scores.
• Benchmarking data quality filtering methods against the original FineWeb2 heuristic baseline.
• Training multilingual language models using a curated subset of web text across 36 languages.

Strengths

• Covers text in 36 different languages.
• Provides model-annotated scores for data quality filtering.
• Uses the original FineWeb2 heuristic method as a performance baseline.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

JQL-AI

Collection Method

Model-annotated subset of the FineWeb2 web-crawled dataset.

Freshness

Last updated 2025-08-07 14:31:33; freshness should be verified.