README.md

metadata

task_categories:
  - text-generation
  - text-classification
dataset_info:
  features:
    - name: instruction
      dtype: string
    - name: response
      dtype: string
    - name: prompt_label
      dtype: string
    - name: response_label
      dtype: string
  splits:
    - name: train
      num_bytes: 225057185
      num_examples: 39787
  download_size: 114338356
  dataset_size: 225057185
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

ThinkSafe Dataset

This dataset is associated with the paper THINKSAFE: Self-Generated Safety Alignment for Reasoning Models.

Paper: https://arxiv.org/abs/2601.23143
GitHub: https://github.com/seanie12/ThinkSafe.git

Dataset Description

This dataset contains 39,787 training examples with instructions and responses, labeled for safety alignment. Each example includes:

• instruction: The input instruction
• response: The generated response
• prompt_label: Safety label for the prompt
• response_label: Safety label for the response

Citation

If you use this dataset, please cite:

@article{lee2025thinksafe,
  title={THINKSAFE: Self-Generated Safety Alignment for Reasoning Models},
  author={Lee, Seanie and others},
  journal={arXiv preprint arXiv:2601.23143},
  year={2025}
}