Machine Unlearning on TOFU Retain95
2.6Refusal RateGuardrail Filter
Evaluation Results
| Method | Links | |
|---|---|---|
| Guardrail FilterBackbone=Llama-2-7B-chat-hf2026.05 | 2.6 | |
| ICCUBackbone=Llama-3-8B-Instruct2026.05 | 3.9 | |
| ICCUBackbone=Llama-2-7B-chat-hf2026.05 | 4.3 | |
| Guardrail ClassifierBackbone=Llama-3-8B-Instruct2026.05 | 4.8 | |
| Guardrail ClassifierBackbone=Llama-2-7B-chat-hf2026.05 | 5 | |
| Guardrail FilterBackbone=Llama-3-8B-Instruct2026.05 | 16.7 |