Unlearning on TOFU (forget05 retain95)
100FSR (%)AN
Evaluation results
| Method | Links | ||
|---|---|---|---|
| ANModel=LLaMA 3.2-3B, Judge=Qwen 2.5-7B2026.05 | 100 | 0.2 | |
| GAModel=LLaMA 3.2-3B, Judge=Qwen 2.5-7B2026.05 | 68 | 32 | |
| MAATModel=LLaMA 3.2-3B, Judge=Qwen 2.5-7B2026.05 | 67.5 | 46.6 | |
| RO-FTModel=LLaMA 3.2-3B, Judge=Qwen 2.5-7B2026.05 | 67 | 45.1 | |
| RO-FTModel=Gemma 3-4B, Judge=Qwen 2.5-7B2026.05 | 63 | 46.2 | |
| ANModel=Gemma 3-4B, Judge=Qwen 2.5-7B2026.05 | 62 | 33.8 | |
| MAATModel=Gemma 3-4B, Judge=Qwen 2.5-7B2026.05 | 61.5 | 48.7 | |
| GA+KLModel=LLaMA 3.2-3B, Judge=Qwen 2.5-7B2026.05 | 57.5 | 39.6 | |
| GA+KLModel=Gemma 3-4B, Judge=Qwen 2.5-7B2026.05 | 54 | 47.4 | |
| GAModel=Gemma 3-4B, Judge=Qwen 2.5-7B2026.05 | 53 | 49.7 |