metadata
dataset_info:
features:
- name: text
dtype: string
- name: meta
struct:
- name: pile_set_name
dtype: string
splits:
- name: train
num_bytes: 542456309
num_examples: 100000
download_size: 202975613
dataset_size: 542456309
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
Dataset Creation Process
These subsets were created by streaming over the rows from the original dataset and filtering by the meta column. Each subset is generally limited to the first 100,000 qualifying rows encountered.
Note: Some subsets contain fewer than 100,000 rows due to the sparsity of certain categories in the original dataset. In these cases, the rows may not be the first encountered but are unevenly distributed due to parallelization during collection.
Future Updates
Future updates may include completing the smaller subsets to reach the 100,000 row target, which could result in breaking changes for those specific subsets.