DCLM-Edu: Filtered Educational Web Text for Small Language Models

DCLM-Edu is a filtered version of the DCLM dataset, annotated for educational quality on a scale from 0 to 5. It contains only samples with a score higher than 2 and was used to train the SmolLM2-135M and SmolLM2-360M models. The dataset was created by HuggingFaceTB and was last updated on March 7, 2025.

Use Cases

• Training small language models based on filtered educational web text.
• Benchmarking educational content quality classifiers based on the 0-5 scoring scale.
• Fine-tuning language models for pedagogical or instructional tasks based on curated web content.
• Studying the impact of data quality filtering on model performance based on the described filtering process.

Strengths

• Data is filtered for educational quality using a classifier scoring from 0 to 5.
• Dataset has been used to train specific models, namely SmolLM2-135M and SmolLM2-360M.
• Last update timestamp is precisely recorded as 2025-03-07 16:24:22.

Limitations

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

Provenance

Source

Filtered from the DCLM dataset using the FineWeb-Edu classifier.

Collection Method

Web pages were annotated for educational quality on a 0-5 scale; only samples with a score >2 were kept.

Freshness

Last updated 2025-03-07 16:24:22.