Machine Unlearning on TOFU Chinese (10% forget)
6.7Refusal Rate (Forget)Guardrail Classifier
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Guardrail ClassifierInput Variant=Chinese2026.05 | 6.7 | 0.6 | |
| ICCUInput Variant=Chinese2026.05 | 90.7 | 27.7 |
Loading the SOTA2 catalog…