Morton-Li's picture
Update README
b790c34 verified
metadata
license: odc-by
task_categories:
  - text-generation
language:
  - en
pretty_name: FineWeb-Edu-Quality4plus
configs:
  - config_name: default
    data_files:
      - split: sample
        path: sample/*
      - split: train
        path: data/*/*
    features:
      - name: text
        dtype: string
      - name: id
        dtype: string
      - name: dump
        dtype: string
      - name: url
        dtype: string
      - name: date
        dtype: string
      - name: file_path
        dtype: string
      - name: language
        dtype: string
      - name: language_score
        dtype: float64
      - name: token_count
        dtype: int64
      - name: score
        dtype: float64
      - name: int_score
        dtype: int64
  - config_name: v1.0.0-v1.4.0
    data_files:
      - split: train
        path: data/v1.0.0-v1.4.0/*

📘 FineWeb-Edu-Quality4plus

Overview

FineWeb-Edu-Quality4plus is a high-quality filtered subset of the original HuggingFaceFW/fineweb-edu dataset (ODC-By License).

This subset retains only samples with:

  • quality_score ≥ 4

The goal is to provide a cleaner and more reliable dataset suitable for language model pre-training, instruction tuning, education-related NLP, and quality-sensitive downstream tasks.

This work is independent and not affiliated with the official FineWeb / FineWeb-Edu maintainers.


Key Features

  • Derived from FineWeb-Edu (ODC-By License)
  • Quality-filtered using quality_score ≥ 4
  • No modification to the original text content
  • Cleaner distribution of high-quality educational web data
  • Suitable for LLM pretraining and fine-tuning

Dataset Construction

Source Dataset

Filtering Rule

quality_score >= 4

Processing Notes

  • The dataset retains exactly the original text and metadata.
  • Only samples failing the quality threshold are removed.
  • No normalization, rewriting, or translation has been applied.

Use Cases

Recommended

  • Pretraining decoder/encoder–decoder language models
  • Instruction tuning
  • Education-domain NLP research
  • Quality-sensitive text modeling
  • Dataset distillation or mixture construction

Not Recommended

  • Tasks requiring the full low-quality distribution of FineWeb-Edu
  • Research requiring unchanged statistical distribution of the full dataset

License

This dataset is distributed under:

  • Open Data Commons Attribution License (ODC-By) v1.0 (same as the source dataset)

The ODC-By license and attribution to the original creators are preserved.


Citation

If you use this dataset, please cite:

@misc{fineweb-edu-quality4plus,
  title  = {FineWeb-Edu-Quality4plus: A High-Quality Subset of FineWeb-Edu},
  author = {Morton Li},
  year   = {2025},
  note   = {Derived from HuggingFaceFW/fineweb-edu (ODC-By License)}
}

Acknowledgements

Special thanks to the creators of:

  • FineWeb-Edu

Their open datasets enable high-quality research for the entire community.