metadata
dataset_info:
features:
- name: question
dtype: string
- name: answer
dtype: string
- name: docid
dtype: string
- name: retrieved_docids
sequence: string
- name: retrieved_doc_scores
sequence: float64
splits:
- name: train
num_bytes: 3912970806
num_examples: 1509096
- name: test
num_bytes: 39528338
num_examples: 15244
download_size: 3285882072
dataset_size: 3952499144
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
- split: test
path: data/test-*
The dataset, introduced within the IACC paper and associated project, is derived from Wikipedia and consists of a long-form question-and-answer collection, encompassing training and test sets.