metadata
dataset_info:
features:
- name: id
dtype: int64
- name: question
dtype: string
- name: pred_answer
dtype: string
- name: topk
struct:
- name: contents
sequence: string
- name: docid
sequence: int64
- name: answer
sequence: string
splits:
- name: train
num_bytes: 1244684721
num_examples: 222849
download_size: 706547692
dataset_size: 1244684721
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
The dataset is originated by dmrau/multi_qa.
The topk passages are retrieved by SPLADE-v3 and the pred_answer is generated with Mistral-7B-Instruct-v0.2 with the retrieved passages.