---
dataset_info:
  features:
  - name: text
    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: metadata
    struct:
    - name: fineweb_dump
      dtype: string
    - name: fineweb_id
      dtype: string
    - name: section
      dtype: string
    - name: url
      dtype: string
  splits:
  - name: train
    num_bytes: 5122834071
    num_examples: 1487373
  download_size: 3119485747
  dataset_size: 5122834071
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: odc-by
language:
- en
tags:
- fineweb
size_categories:
- 1M<n<10M
---
This is a subset of the [Fineweb](https://huggingface.co/datasets/HuggingFaceFW/fineweb) dataset trimmed down to approximately one billion tokens. No special frills. We sampled from the 10 billion token subset to create this one.