metadata
dataset_info:
features:
- name: idx
dtype: int64
- name: img_url
dtype: string
- name: image_tokens
sequence: int64
- name: caption
dtype: string
- name: clean_caption
dtype: string
- name: input_ids
sequence: int32
- name: attention_mask
sequence: int8
splits:
- name: train
num_bytes: 944695861
num_examples: 527187
download_size: 174001714
dataset_size: 944695861
configs:
- config_name: default
data_files:
- split: train
path: data/train-*