📖 unlearn_dataset

The unlearn_dataset serves as a benchmark for evaluating unlearning methodologies in pre-trained large language models across diverse domains, including arXiv, GitHub.

🔍 Loading the datasets

To load the dataset:

from datasets import load_dataset

dataset = load_dataset("llmunlearn/unlearn_dataset", name="arxiv", split="forget")

• Available configuration names and corresponding splits:
  • arxiv: forget, approximate, retain
  • github: forget, approximate, retain
  • general: evaluation, retain

🛠️ Codebase

For evaluating unlearning methods on our datasets, visit our GitHub repository.

⭐ Citing our Work

If you find our codebase or dataset useful, please consider citing our paper:

@article{yao2024machine,
  title={Machine Unlearning of Pre-trained Large Language Models},
  author={Yao, Jin and Chien, Eli and Du, Minxin and Niu, Xinyao and Wang, Tianhao and Cheng, Zezhou and Yue, Xiang},
  journal={arXiv preprint arXiv:2402.15159},
  year={2024}
}