metadata

dataset_info:
  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
  splits:
    - name: train
      num_bytes: 51073574.99634302
      num_examples: 14874
  download_size: 30919243
  dataset_size: 51073574.99634302
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: odc-by
task_categories:
  - text-generation
language:
  - en
size_categories:
  - 10K<n<100K

FineWeb 10MT

Roughly 10 million tokens worth of English FineWeb data. Obtained by taking the 10BT Sample of FineWeb, shuffling it, sharding into 1000 shards, and selecting the first shard.

Code

To reproduce FineWeb 10MT, simply use the following:

from datasets import load_dataset

fineweb = load_dataset("HuggingFaceFW/fineweb", "sample-10BT", split="train").shuffle().shard(num_shards=1000, index=0)

token = "token"
fineweb.push_to_hub("OxxoCodes/fineweb-10MT", token=token)