README.md

metadata

dataset_info:
  features:
    - name: text
      dtype: string
  splits:
    - name: train
      num_bytes: 1103636977
      num_examples: 861992
    - name: validation
      num_bytes: 11037684
      num_examples: 8544
  download_size: 555101925
  dataset_size: 1114674661
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
license: mit
task_categories:
  - text-generation
language:
  - en

Dataset Description

This dataset is a cleaned and reformatted version of roneneldan/TinyStoriesInstruct.

The following changed were made to improve data quality and make it more suitable for instruction-tuning:

• Filtered repetitive openings: Samples with repeated phrases like "once upon a time" at the beginning were removed.
• Removed empty or invalid entries: Any stories with missing or empty content were excluded.
• Standardized instruction-response format: Each example is converted into the following structure:

  ### Instruction:
  {instruction}
  
  ### Response:
  {response}{tokenizer.eos_token}