metadata
dataset_info:
features:
- name: text
dtype: string
- name: pile_idx
dtype: int64
splits:
- name: train
num_bytes: 5213971245
num_examples: 842967
- name: validation
num_bytes: 2658704
num_examples: 500
- name: test
num_bytes: 63271721
num_examples: 10000
download_size: 2903588242
dataset_size: 5279901670
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: validation
path: data/validation-*
- split: test
path: data/test-*
language:
- en
Produced with the MiniCorpus pipeline, a reproduction and investigation of the MiniPile data-distillation method (Kaddour, 2023).
This dataset is derived from and based on the contents of The Pile Deduplicated.