metadata
dataset_info:
features:
- name: text
dtype: string
- name: pile_idx
dtype: int64
splits:
- name: train
num_bytes: 6250106397
num_examples: 999998
- name: validation
num_bytes: 62229560
num_examples: 9999
- name: test
num_bytes: 3059198
num_examples: 501
download_size: 3568305514
dataset_size: 6315395155
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.