Watermark Spoofing on EWD (test)
56.5SSRRLSpoofer
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| RLSpooferModel=Qwen3-4B2026.04 | 56.5 | 87 | 73 | 4.88 | 6.68 | |
| DITTOModel=Qwen3-4B2026.04 | 56 | 94.3 | 68 | 5.62 | 6.66 | |
| DistillModel=Qwen2.5-3B-Instruct2026.04 | 55.5 | 81.5 | 76 | 4.4 | 7.31 | |
| RLSpooferModel=Llama3.2-3B-Instruct2026.04 | 54.5 | 79.5 | 70 | 4.99 | 6.79 | |
| RLSpooferModel=Qwen3-0.6B2026.04 | 54.3 | 80.5 | 73 | 5.36 | 5.77 | |
| DistillModel=Llama3.2-3B-Instruct2026.04 | 53.8 | 71.3 | 77 | 4.49 | 7.22 | |
| RLSpooferModel=Qwen3-1.7B2026.04 | 53.5 | 68.5 | 76 | 5.15 | 6.5 | |
| RLSpooferModel=Qwen2.5-3B-Instruct2026.04 | 53.5 | 83 | 70 | 4.66 | 6.8 | |
| DistillModel=Qwen3-4B2026.04 | 51.3 | 62.5 | 81 | 4.75 | 7.66 | |
| DistillModel=Qwen3-1.7B2026.04 | 43.8 | 53.8 | 80 | 4.59 | 7.27 | |
| DistillModel=Qwen3-0.6B2026.04 | 42.3 | 57.3 | 75 | 4.5 | 6.62 | |
| DITTOModel=Qwen3-1.7B2026.04 | 21.5 | 96 | 53 | 5.26 | 4.68 | |
| DITTOModel=Llama3.2-3B-Instruct2026.04 | 19.3 | 99.5 | 54 | 4.9 | 5.13 | |
| DITTOModel=Qwen2.5-3B-Instruct2026.04 | 14 | 99.5 | 50 | 4.95 | 5.32 | |
| DITTOModel=Qwen3-0.6B2026.04 | 7.5 | 97.3 | 43 | 4.78 | 3 | |
| DPOModel=Llama3.2-3B-Instruct2026.04 | 2.5 | 1.25 | 49 | 2 | 2.62 | |
| DPOModel=Qwen3-0.6B2026.04 | 0.25 | 0.75 | 57 | 1.95 | 4.42 | |
| DPOModel=Qwen3-1.7B2026.04 | 0.25 | 0.25 | 94 | 2.35 | 8.17 | |
| DPOModel=Qwen3-4B2026.04 | 0.25 | 0.5 | 78 | 2.24 | 7.55 | |
| DPOModel=Qwen2.5-3B-Instruct2026.04 | 0 | 0.75 | 88 | 2 | 7.76 |