TOFU: Task of Fictitious Unlearning

The TOFU dataset serves as a benchmark for evaluating unlearning performance of large language models on realistic tasks.