Machine Unlearning on TOFU Retain90
0Refusal RateGuardrail Classifier
Evaluation Results
| Method | Links | |
|---|---|---|
| Guardrail ClassifierBackbone=Llama-3-8B-Instruct2026.05 | 0 | |
| Guardrail ClassifierBackbone=Llama-2-7B-chat-hf2026.05 | 0 | |
| ICCUBackbone=Llama-3-8B-Instruct2026.05 | 0.6 | |
| ICCUBackbone=Llama-2-7B-chat-hf2026.05 | 0.9 | |
| Guardrail FilterBackbone=Llama-2-7B-chat-hf2026.05 | 1.8 | |
| Guardrail FilterBackbone=Llama-3-8B-Instruct2026.05 | 14.8 |