metadata
dataset_info:
features:
- name: id
dtype: string
- name: embedding
sequence: float16
length: 1024
splits:
- name: corpus
num_bytes: 36671273340
num_examples: 17801589
- name: test_query
num_bytes: 133575
num_examples: 65
download_size: 35082105584
dataset_size: 36671406915
configs:
- config_name: default
data_files:
- split: corpus
path: data/corpus-*
- split: test_query
path: data/test_query-*
stella_en_1.5B_v5 embeddings of TREC24 BioGen PubMed corpus and test queries.
Corpus contains unique PMIDs from TREC BioGen: 20723868 samples
Then further removed the samples that have empty abstracts: 17801589 samples
The corpus input text for Stella encoder model is title + abstract (space separator)
The query prompt input text for Stella encoder is Instruct: Given a medical query, retrieve documents that answer the query.\nQuery: {query}
DatasetDict({
corpus: Dataset({
features: ['id', 'embedding'],
num_rows: 17801589
})
test_query: Dataset({
features: ['id', 'embedding'],
num_rows: 65
})
})