metadata
license: unknown
dataset_info:
- config_name: corpus
features:
- name: query_id
dtype: string
- name: snippets
dtype: string
- name: air_date
dtype: string
- name: category
dtype: string
- name: value
dtype: string
- name: round
dtype: string
- name: show_number
dtype: int32
- name: doc_id
dtype: string
- name: __index_level_0__
dtype: int64
splits:
- name: train
num_bytes: 6252715344
num_examples: 14120776
download_size: 3271155810
dataset_size: 6252715344
- config_name: qa_data
features:
- name: query_id
dtype: string
- name: question
dtype: string
- name: answer
dtype: string
- name: search_results
struct:
- name: related_links
sequence: string
- name: snippets
sequence: string
- name: titles
sequence: string
- name: urls
sequence: string
- name: doc_id
sequence: string
- name: __index_level_0__
dtype: int64
splits:
- name: train
num_bytes: 6503932619
num_examples: 173397
- name: test
num_bytes: 1830028629
num_examples: 43350
download_size: 5008413626
dataset_size: 8333961248
configs:
- config_name: corpus
data_files:
- split: train
path: corpus/train-*
- config_name: qa_data
data_files:
- split: train
path: qa_data/train-*
- split: test
path: qa_data/test-*
preprocessed_SearchQA
The SearchQA question-answer pairs originate from J! Archive2, which comprehensively archives all question-answer pairs from the renowned television show Jeopardy! The passages, sourced from Google search web page snippets. We offer passage metadata, encompassing details like 'air_date,' 'category,' 'value,' 'round,' and 'show_number,' enabling you to enhance retrieval performance at your discretion. Should you require further details about SearchQA, please refer to below links.
The dataset is derived from searhQA.
This preprocessed dataset is for RAG. For more information about our task, visit our repository!
Preprocess SearchQA dataset code for RAG benchmark.
More information, refer to this link! huggingface