LLM Watermark Spoofing on SWEET
54.5SSRRLSpoofer
Evaluation results
| Method | Links | ||
|---|---|---|---|
| RLSpooferModel=Qwen2.5-3B-Instruct2026.04 | 54.5 | 75 | |
| RLSpooferModel=Llama3.2-3B-Instruct2026.04 | 54.5 | 74 | |
| RLSpooferModel=Qwen3-4B2026.04 | 52.3 | 75 | |
| RLSpooferModel=Qwen3-1.7B2026.04 | 52 | 71 | |
| RLSpooferModel=Qwen3-0.6B2026.04 | 50.5 | 79 | |
| DistillModel=Qwen2.5-3B-Instruct2026.04 | 49.8 | 77 | |
| DistillModel=Llama3.2-3B-Instruct2026.04 | 45.5 | 76 | |
| DITTOModel=Qwen3-4B2026.04 | 43.3 | 66 | |
| DistillModel=Qwen3-4B2026.04 | 37.3 | 82 | |
| DITTOModel=Qwen3-1.7B2026.04 | 29 | 56 | |
| DistillModel=Qwen3-1.7B2026.04 | 26.8 | 79 | |
| DITTOModel=Llama3.2-3B-Instruct2026.04 | 24.3 | 56 | |
| DITTOModel=Qwen2.5-3B-Instruct2026.04 | 22.3 | 54 | |
| DistillModel=Qwen3-0.6B2026.04 | 20 | 76 | |
| DITTOModel=Qwen3-0.6B2026.04 | 6.75 | 41 | |
| DPOModel=Llama3.2-3B-Instruct2026.04 | 0.5 | 53 | |
| DPOModel=Qwen3-0.6B2026.04 | 0 | 76 | |
| DPOModel=Qwen3-1.7B2026.04 | 0 | 84 | |
| DPOModel=Qwen3-4B2026.04 | 0 | 78 | |
| DPOModel=Qwen2.5-3B-Instruct2026.04 | 0 | 87 |