Collision Attack on Input Reconstruction (First split)
1BERTScoreOriginal
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| OriginalModel=LLaMA-7B2025.08 | 1 | 1 | |
| OriginalModel=LLaMA-3.2-1B2025.08 | 1 | 0.994 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.2-1B2025.08 | 0.973 | 0.966 | |
| OriginalModel=LLaMA-3.1-8B-Distilled2025.08 | 0.885 | 0.858 | |
| OriginalModel=LLaMA-3.2-1B2025.08 | 0.877 | 0.709 | |
| OriginalModel=LLaMA-3.2-1B, Protect Type=Original2025.08 | 0.877 | 0.709 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.2-1B2025.08 | 0.849 | 0.604 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.2-1B, Protect Type=(10^8, 10^-5)-DP2025.08 | 0.849 | 0.604 | |
| OriginalModel=LLaMA-3.1-8B-Distilled2025.08 | 0.642 | 0.633 | |
| OriginalModel=LLaMA-3.1-8B-Distilled, Protect Type=Original2025.08 | 0.642 | 0.633 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.1-8B-Distilled2025.08 | 0.526 | 0.53 | |
| OriginalModel=LLaMA-7B2025.08 | 0.449 | 0.5 | |
| OriginalModel=LLaMA-7B, Protect Type=Original2025.08 | 0.449 | 0.5 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.1-8B-Distilled2025.08 | 0.32 | 0.291 | |
| (10^8, 10^-5)-DPModel=LLaMA-3.1-8B-Distilled, Protect Type=(10^8, 10^-5)-DP2025.08 | 0.32 | 0.291 | |
| (10^8, 10^-5)-DPModel=LLaMA-7B2025.08 | 0.109 | 0.097 | |
| (10^8, 10^-5)-DPModel=LLaMA-7B2025.08 | 0.082 | 0.041 | |
| (10^8, 10^-5)-DPModel=LLaMA-7B, Protect Type=(10^8, 10^-5)-DP2025.08 | 0.082 | 0.041 | |
| KV-CloakModel=LLaMA-3.2-1B2025.08 | 0.072 | 0 | |
| KV-CloakModel=LLaMA-3.2-1B, Protect Type=KV-Cloak2025.08 | 0.072 | 0 | |
| KV-CloakModel=LLaMA-7B2025.08 | 0.07 | 0 | |
| KV-CloakModel=LLaMA-3.1-8B-Distilled2025.08 | 0.07 | 0 | |
| KV-CloakModel=LLaMA-7B, Protect Type=KV-Cloak2025.08 | 0.07 | 0 | |
| KV-CloakModel=LLaMA-3.1-8B-Distilled, Protect Type=KV-Cloak2025.08 | 0.07 | 0 | |
| KV-CloakModel=LLaMA-3.2-1B2025.08 | 0.051 | 0.002 | |
| KV-CloakModel=LLaMA-3.1-8B-Distilled2025.08 | 0.041 | 0.003 | |
| KV-CloakModel=LLaMA-7B2025.08 | 0.036 | 0.044 |