ToFU

TOFU: Task of Fictitious Unlearning 🍢 The TOFU dataset serves as a benchmark for evaluating unlearning performance of large language models on realistic tasks. The dataset comprises question-answer pairs based on autobiographies of 200 different authors that do not exist and are completely fictitiously generated by th

Papers using ToFU (14)

Less is More: Geometric Unlearning for LLMs with Minimal Data Disclosure2026BalDRO: A Distributionally Robust Optimization based Framework for Large Language Model Unlearning2026Leak@$k$: Unlearning Does Not Make LLMs Forget Under Probabilistic Decoding2025Hierarchical Federated Unlearning for Large Language Models2025LLM Unlearning using Gradient Ratio-Based Influence Estimation and Noise Injection2025GUARD: Guided Unlearning and Retention via Data Attribution for Large Language Models2025UCD: Unlearning in LLMs via Contrastive Decoding2025OpenUnlearning: Accelerating LLM Unlearning via Unified Benchmarking of Methods and Metrics2025Constrained Entropic Unlearning: A Primal-Dual Framework for Large Language Models2025OBLIVIATE: Robust and Practical Machine Unlearning for Large Language Models2025UniErase: Towards Balanced and Precise Unlearning in Language Models2025UIPE: Enhancing LLM Unlearning by Removing Knowledge Related to Forgetting Targets2025Soft Token Attacks Cannot Reliably Audit Unlearning in Large Language Models2025Answer When Needed, Forget When Not: Language Models Pretend to Forget via In-Context Knowledge Unlearning2024