Unlearning on Syntax-preserving dataset (forget set)
0.51PrivLeakGA
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GAModel=Qwen-7B, Forget sets=10%, Epoch=62025.12 | 0.51 | 0.47 | |
| SimNPOModel=Qwen-7B, Forget sets=10%, Epoch=52025.12 | 0.54 | 0.53 | |
| FiFSLModel=DeepSeek-7B, Forget sets=30%, Epoch=12025.12 | 0.55 | 0.45 | |
| FiFSLModel=DeepSeek-7B, Forget sets=10%, Epoch=12025.12 | 0.56 | 0.47 | |
| FiFSLModel=DeepSeek-7B, Forget sets=20%, Epoch=12025.12 | 0.56 | 0.47 | |
| FiFSLModel=Qwen-7B, Forget sets=30%, Epoch=12025.12 | 0.56 | 0.49 | |
| FiFSLModel=Qwen-7B, Forget sets=20%, Epoch=12025.12 | 0.57 | 0.51 | |
| FiFSLModel=Llama-8B, Forget sets=30%, Epoch=12025.12 | 0.6 | 0.51 | |
| FiFSLModel=Qwen-7B, Forget sets=10%, Epoch=12025.12 | 0.6 | 0.47 | |
| SimNPOModel=Qwen-7B, Forget sets=20%, Epoch=42025.12 | 0.6 | 0.5 | |
| SimNPOModel=Llama-8B, Forget sets=10%, Epoch=32025.12 | 0.61 | 0.5 | |
| GAModel=Qwen-7B, Forget sets=20%, Epoch=42025.12 | 0.61 | 0.51 | |
| FiFSLModel=Llama-8B, Forget sets=20%, Epoch=12025.12 | 0.64 | 0.51 | |
| GAModel=Qwen-7B, Forget sets=30%, Epoch=32025.12 | 0.64 | 0.5 | |
| SimNPOModel=Qwen-7B, Forget sets=30%, Epoch=32025.12 | 0.64 | 0.5 | |
| FiFSLModel=Llama-7B, Forget sets=20%, Epoch=22025.12 | 0.65 | 0.54 | |
| FiFSLModel=Llama-7B, Forget sets=30%, Epoch=12025.12 | 0.65 | 0.53 | |
| GAModel=Llama-8B, Forget sets=30%, Epoch=22025.12 | 0.65 | 0.5 | |
| SimNPOModel=Llama-8B, Forget sets=30%, Epoch=22025.12 | 0.65 | 0.5 | |
| GAModel=Llama-8B, Forget sets=10%, Epoch=22025.12 | 0.66 | 0.58 | |
| GAModel=Llama-8B, Forget sets=20%, Epoch=22025.12 | 0.67 | 0.5 | |
| FiFSLModel=Llama-8B, Forget sets=10%, Epoch=12025.12 | 0.68 | 0.49 | |
| SimNPOModel=Llama-8B, Forget sets=20%, Epoch=22025.12 | 0.69 | 0.5 | |
| FiFSLModel=Llama-7B, Forget sets=10%, Epoch=12025.12 | 0.7 | 0.53 | |
| GAModel=DeepSeek-7B, Forget sets=30%, Epoch=52025.12 | 0.71 | 0.51 | |
| SimNPOModel=DeepSeek-7B, Forget sets=30%, Epoch=52025.12 | 0.71 | 0.5 | |
| SimNPOModel=Llama-7B, Forget sets=30%, Epoch=42025.12 | 0.72 | 0.53 | |
| SimNPOModel=DeepSeek-7B, Forget sets=20%, Epoch=62025.12 | 0.72 | 0.51 | |
| GAModel=Llama-7B, Forget sets=30%, Epoch=42025.12 | 0.73 | 0.54 | |
| GAModel=DeepSeek-7B, Forget sets=20%, Epoch=62025.12 | 0.73 | 0.52 | |
| GAModel=Llama-7B, Forget sets=10%, Epoch=62025.12 | 0.75 | 0.52 | |
| SimNPOModel=Llama-7B, Forget sets=10%, Epoch=62025.12 | 0.75 | 0.51 | |
| SimNPOModel=DeepSeek-7B, Forget sets=10%, Epoch=72025.12 | 0.75 | 0.5 | |
| GAModel=DeepSeek-7B, Forget sets=10%, Epoch=72025.12 | 0.77 | 0.52 | |
| GAModel=Llama-7B, Forget sets=20%, Epoch=42025.12 | 0.82 | 0.53 | |
| SimNPOModel=Llama-7B, Forget sets=20%, Epoch=42025.12 | 0.82 | 0.53 | |
| Qwen-7B BaseModel=Qwen-7B, Forget sets=0%2025.12 | 0.87 | 0.96 | |
| Llama-8B BaseModel=Llama-8B, Forget sets=0%2025.12 | 0.96 | 0.97 | |
| DeepSeek-7B BaseModel=DeepSeek-7B, Forget sets=0%2025.12 | 0.97 | 0.83 | |
| Llama-7B BaseModel=Llama-7B, Forget sets=0%2025.12 | 0.98 | 0.8 |