README.md

metadata

language:
  - fi
license: cc-by-sa-3.0
task_categories:
  - text-generation
dataset_info:
  features:
    - name: instruction
      dtype: string
    - name: response
      dtype: string
    - name: instruction_orig
      dtype: string
    - name: response_orig
      dtype: string
    - name: response_len
      dtype: int64
    - name: rank
      dtype: float64
    - name: __index_level_0__
      dtype: int64
    - name: response_orig_grade
      dtype: string
    - name: response_judgelm
      dtype: string
  splits:
    - name: train
      num_bytes: 3272561
      num_examples: 2160
  download_size: 1872337
  dataset_size: 3272561
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Dataset Card for Finnish-NLP/boolq-deepl-translated-sft

Creation process

• Load data from google/boolq translated with deepl
• Write rows to .txt file with *** on a newline separating instruction/response and then END on a newline separating samples
• Upload file to deepl.com for file translation --> parse samples back from translated files --> Maybe some additional cleaning/filtering based on fasttext langdetect / kenlm perplexity