README.md

metadata

dataset_info:
  features:
    - name: instruction
      dtype: string
    - name: response
      dtype: string
  splits:
    - name: train
      num_bytes: 131141
      num_examples: 410
  download_size: 78634
  dataset_size: 131141
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
license: cc-by-sa-4.0
task_categories:
  - text-generation
language:
  - fi
tags:
  - SFT

Dataset Card for Finnish-NLP/ai2arc-deepl-translated-sft

Creation process

• Load data from allenai/ai2_arc translated with deepl
• Do zero shot classification with facebook/bart-large-mnli with the following prompt:

preds =  pipe(f'{row["input"]} is a question about:', candidate_labels=["USA related question", "Math related question", "General question", "Coding related question"])

• Filter out rows with too high scores in following categories ["USA related question", "Math related question","Coding related question"]
• 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