README.md

metadata

dataset_info:
  features:
    - name: chosen
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: rejected
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
    - name: prompt
      list:
        - name: content
          dtype: string
        - name: role
          dtype: string
  splits:
    - name: train
      num_bytes: 35373088
      num_examples: 12859
  download_size: 19079320
  dataset_size: 35373088
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: apache-2.0

Attribution

Formatted from Intel/orca_dpo_pairs for DPO training and you only need to apply tokenizer chat template.

License

This dataset is licensed under Apache-2.0.