This subset provides a standardized benchmark for evaluating sparse model performance on HotpotQA data, with negative examples pooled using BM25 retrieval.
This subset was created using the make_beir_subset.py script with the following command:
python ./make_beir_subset.py --dataset hotpotqa --es-host
http://localhost:9200 --force-reindex --split test
Dataset: hotpotqa
Split: test (test set)
ES Host:… See the full description on the dataset page:
https://huggingface.co/datasets/freethenation/hotpotqa-sub.
