metadata
dataset_info:
features:
- name: input_ids
sequence: int32
splits:
- name: train
num_bytes: 1076886272
num_examples: 27332
download_size: 487651276
dataset_size: 1076886272
task_categories:
- text-generation
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
This dataset contains data used in the paper From Hours to Minutes: Lossless Acceleration of Ultra Long Sequence Generation up to 100K Tokens.