multiqa_top5 / README.md
bgchoi's picture
Update README.md
3c895d4 verified
|
Raw
History Blame Contribute Delete
712 Bytes
metadata
dataset_info:
  features:
    - name: id
      dtype: int64
    - name: question
      dtype: string
    - name: pred_answer
      dtype: string
    - name: topk
      struct:
        - name: contents
          sequence: string
        - name: docid
          sequence: int64
    - name: answer
      sequence: string
  splits:
    - name: train
      num_bytes: 1244684721
      num_examples: 222849
  download_size: 706547692
  dataset_size: 1244684721
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

The dataset is originated by dmrau/multi_qa.

The topk passages are retrieved by SPLADE-v3 and the pred_answer is generated with Mistral-7B-Instruct-v0.2 with the retrieved passages.