metadata

dataset_info:
  features:
    - name: text
      dtype: string
    - name: id
      dtype: string
    - name: metadata
      struct:
        - name: dump
          dtype: string
        - name: url
          dtype: string
        - name: date
          dtype: timestamp[ms]
        - 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
  splits:
    - name: train
      num_bytes: 223818106668.24948
      num_examples: 43368339
    - name: test
      num_bytes: 22387828.750528496
      num_examples: 4338
  download_size: 129908705334
  dataset_size: 223840494497
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: test
        path: data/test-*
license: odc-by
task_categories:
  - text-generation
language:
  - en
size_categories:
  - 10M<n<100M

Fineweb-edu-dedup 45B

This dataset was filtered from HuggingFaceTB/smollm-corpus. We selected the fineweb-edu and further filtered with score > 3.0 to make the dataset with higher quality; there are 45B GPT2 tokens in this dataset.

Acknowledgement

We appreciate the efforts from HuggingFaceTB team to release these high-quality dataset and facilitate LLM community