Watermark Detectability on C4-RealNewsLike Dipper-1 (test)
0.977AUCMarkTune
Evaluation results
| Method | Links | ||
|---|---|---|---|
| MarkTuneBackbone=Qwen3-4B, Temperature=0.72025.12 | 0.977 | 0.852 | |
| MarkTuneBackbone=Llama2-7B, Temperature=1.02025.12 | 0.961 | 0.836 | |
| Gumbel-maxBackbone=Llama2-7B, Temperature=1.02025.12 | 0.95 | 0.7 | |
| Gumbel-maxBackbone=Qwen3-4B, Temperature=0.72025.12 | 0.945 | 0.748 | |
| SynthIDBackbone=Llama2-7B, Temperature=1.02025.12 | 0.943 | 0.552 | |
| SynthIDBackbone=Qwen3-4B, Temperature=0.72025.12 | 0.941 | 0.576 | |
| GaussMarkBackbone=Qwen3-4B, Temperature=0.72025.12 | 0.915 | 0.438 | |
| KGW-2Backbone=Qwen3-4B, Temperature=0.72025.12 | 0.884 | 0.372 | |
| KGW-2Backbone=Llama2-7B, Temperature=1.02025.12 | 0.862 | 0.322 | |
| GaussMarkBackbone=Llama2-7B, Temperature=1.02025.12 | 0.857 | 0.24 | |
| KGW-DBackbone=Qwen3-4B, Temperature=0.72025.12 | 0.855 | 0.274 | |
| KGW-DBackbone=Llama2-7B, Temperature=1.02025.12 | 0.814 | 0.238 | |
| KGW-1Backbone=Qwen3-4B, Temperature=0.72025.12 | 0.771 | 0.142 | |
| KGW-1Backbone=Llama2-7B, Temperature=1.02025.12 | 0.757 | 0.116 |