metadata

language:
  - en
multilinguality:
  - monolingual
size_categories:
  - 1M<n<10M
pretty_name: MS MARCO corpus
dataset_info:
  - config_name: passage
    features:
      - name: pid
        dtype: int64
      - name: text
        dtype: string
    splits:
      - name: train
        num_bytes: 3089201649
        num_examples: 8841823
    download_size: 1688656108
    dataset_size: 3089201649
  - config_name: query
    features:
      - name: qid
        dtype: int64
      - name: text
        dtype: string
    splits:
      - name: train
        num_bytes: 48033044
        num_examples: 1010916
    download_size: 34858846
    dataset_size: 48033044
configs:
  - config_name: passage
    data_files:
      - split: train
        path: passage/train-*
  - config_name: query
    data_files:
      - split: train
        path: queries/train-*

MS MARCO Corpus

This dataset allows for a convenient mapping from MS MARCO query/passage ID to the query/passage text. This passage corpus was downloaded from https://msmarco.z22.web.core.windows.net/msmarcoranking/collection.tar.gz, and the queries from https://msmarco.blob.core.windows.net/msmarcoranking/queries.tar.gz (via Wayback Machine).

Usage

This dataset was designed to allow you to perform the following:

from datasets import load_dataset

query_dataset = load_dataset("sentence-transformers/msmarco-corpus", "query", split="train")
qid_to_query = dict(zip(query_dataset["qid"], query_dataset["text"]))
print(qid_to_query[571018])
# => "what are the liberal arts?"

passage_dataset = load_dataset("sentence-transformers/msmarco-corpus", "passage", split="train")
pid_to_passage = dict(zip(passage_dataset["pid"], passage_dataset["text"]))
print(pid_to_passage[7349777])
# => "liberal arts. 1. the academic course of instruction at a college intended to provide general knowledge and comprising the arts, humanities, natural sciences, and social sciences, as opposed to professional or technical subjects."

Related Datasets

This dataset is used for the query and passage texts in the following datasets containing MS MARCO with mined hard negatives.