ar_mmarco_queries / README.md
akhooli's picture
Update README.md
c929dff verified
metadata
license: mit
dataset_info:
  features:
    - name: query_id
      dtype: int64
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 55084982
      num_examples: 808728
  download_size: 33178859
  dataset_size: 55084982
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Note: [Oct. 3, 2024] this dataset is likely not fully aligned in terms of translation or the score are not compatible with Arabic translation. It is being checked and so far, one clear mistranslation was spotted.

This is the Arabic (translated) queries of the MS-MARCO dataset that have scores in this dataset. This dataset is used with akhooli/ar_mmarco_scores and akhooli/ar_mmarco_docs to train ColBERT knowledge distillation models (information retrieval).