metadata
dataset_info:
features:
- name: input_ids
sequence: int32
- name: attention_mask
sequence: int8
- name: position_ids
sequence: int64
- name: loss_mask
sequence: int64
- name: token_entropy
sequence: float64
- name: token_loss
sequence: float64
splits:
- name: train
num_bytes: 152557888
num_examples: 4024
download_size: 10119339
dataset_size: 152557888
configs:
- config_name: default
data_files:
- split: train
path: data/train-*