Watermark Spoofing on SWEET (test)
54.5SSRRLSpoofer
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLSpooferModel=Qwen2.5-3B-Instruct2026.04 | 54.5 | 68.8 | 75 | 6.64 | 6.57 | |
| RLSpooferModel=Llama3.2-3B-Instruct2026.04 | 54.5 | 79.5 | 74 | 6.15 | 6.43 | |
| RLSpooferModel=Qwen3-4B2026.04 | 52.3 | 71.3 | 75 | 5.38 | 6.78 | |
| RLSpooferModel=Qwen3-1.7B2026.04 | 52 | 72 | 71 | 5.92 | 6.45 | |
| RLSpooferModel=Qwen3-0.6B2026.04 | 50.5 | 66.3 | 79 | 4.93 | 5.77 | |
| DistillModel=Qwen2.5-3B-Instruct2026.04 | 49.8 | 67.5 | 77 | 4.37 | 7.31 | |
| DistillModel=Llama3.2-3B-Instruct2026.04 | 45.5 | 62 | 76 | 4.37 | 7.22 | |
| DITTOModel=Qwen3-4B2026.04 | 43.3 | 94.5 | 66 | 5.25 | 6.66 | |
| DistillModel=Qwen3-4B2026.04 | 37.3 | 43.8 | 82 | 4.79 | 7.66 | |
| DITTOModel=Qwen3-1.7B2026.04 | 29 | 92.8 | 56 | 5.22 | 4.68 | |
| DistillModel=Qwen3-1.7B2026.04 | 26.8 | 37.8 | 79 | 4.63 | 7.27 | |
| DITTOModel=Llama3.2-3B-Instruct2026.04 | 24.3 | 100 | 56 | 4.76 | 5.13 | |
| DITTOModel=Qwen2.5-3B-Instruct2026.04 | 22.3 | 99.8 | 54 | 4.78 | 5.32 | |
| DistillModel=Qwen3-0.6B2026.04 | 20 | 33.8 | 76 | 4.68 | 6.62 | |
| DITTOModel=Qwen3-0.6B2026.04 | 6.75 | 95.8 | 41 | 4.52 | 3 | |
| DPOModel=Llama3.2-3B-Instruct2026.04 | 0.5 | 3 | 53 | 1.86 | 2.62 | |
| DPOModel=Qwen3-0.6B2026.04 | 0 | 0 | 76 | 2.71 | 4.42 | |
| DPOModel=Qwen3-1.7B2026.04 | 0 | 0 | 84 | 2.46 | 8.17 | |
| DPOModel=Qwen3-4B2026.04 | 0 | 0 | 78 | 2.02 | 7.55 | |
| DPOModel=Qwen2.5-3B-Instruct2026.04 | 0 | 0.5 | 87 | 1.94 | 7.76 |