Fineweb Edu 1B is a 1-billion-token sample of curated educational web resources, created by author 'codelion' using reservoir sampling. It was last updated in November 2025. The dataset is derived from the HuggingFaceFW/fineweb-edu source.
Use Cases
- Benchmarking language model perplexity on educational text from the Fineweb source.
- Analyzing the vocabulary and topic distribution within the 1B token sample for curriculum insights.
- Training or fine-tuning smaller models on a representative subset of educational web data.
- Conducting ablation studies on model performance by comparing results from this sample against the full dataset.
Strengths
- Contains 1 billion tokens of educational text.
- Created using reservoir sampling, a statistically unbiased random sampling method.
Limitations
- Specific row count, column structure, and file formats are unknown.
- The sample's representativeness of the full dataset's characteristics cannot be verified without the original source statistics.
Provenance
- Source
- HuggingFaceFW/fineweb-edu
- Collection Method
- Reservoir sampling.
- Time Range
- null
- Freshness
- null
- Geography
- null