dataset_info:
- config_name: documents
features:
- name: document
dtype: string
splits:
- name: full
num_bytes: 1263393
num_examples: 3200
download_size: 782809
dataset_size: 1263393
- config_name: questions
features:
- name: question
dtype: string
- name: ground_truth
dtype: string
splits:
- name: full
num_bytes: 73702
num_examples: 918
- name: mini_100
num_bytes: 8028.5403050108935
num_examples: 100
download_size: 56533
dataset_size: 81730.54030501089
configs:
- config_name: documents
data_files:
- split: full
path: documents/full-*
- config_name: questions
data_files:
- split: full
path: questions/full-*
- split: mini_100
path: questions/mini_100-*
RAGU eval dataset for Wikipedia
ragu-mini-wikipedia is a dataset to evaluate RAG pipelines. It consists out of ~900 question and ground truth answers from Wikipedia articles.
In addition to the questions it has a second config with documents for retrieval. The dataset can be used by "indexing" the documents and then using the question and ground truth to evaluate your pipeline.
The dataset is derived from Question-Answer Dataset and slightly modified to rag-datasets/mini_wikipedia.
This modified dataset is distributed under the CC BY-SA 3.0 license, consistent with the licensing of the source datasets.
Acknowledgements
These data were collected by Noah Smith, Michael Heilman, Rebecca Hwa, Shay Cohen, Kevin Gimpel, and many students at Carnegie Mellon University and the University of Pittsburgh between 2008 and 2010. It is released here under CC BY_SA 3.0. Please cite this paper if you write any papers involving the use of the data above:
Smith, N. A., Heilman, M., & Hwa, R. (2008, September). Question generation as a competitive undergraduate course project. In Proceedings of the NSF Workshop on the Question Generation Shared Task and Evaluation Challenge.
TODO: Add blog on how to use it