Watermark Detection on GSM8K
100True Detection Rate (TD)He et al.
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| He et al.Student Model=Llama-3.1-8B, K=202026.02 | 100 | 100 | |
| He et al.Student Model=Llama-3.2-3B, K=202026.02 | 100 | 100 | |
| He et al.Student Model=Qwen2.5-1.5B, K=202026.02 | 100 | 99 | |
| VIAStudent Model=Llama-3.1-8B, K=202026.02 | 100 | 2 | |
| VIAStudent Model=Qwen2.5-1.5B, K=202026.02 | 100 | 0 | |
| OPTStudent Model=Llama-3.1-8B, K=52026.02 | 100 | 0 | |
| OPTStudent Model=Llama-3.1-8B, K=202026.02 | 100 | 2 | |
| OPTStudent Model=Qwen2.5-1.5B, K=202026.02 | 100 | 0 | |
| He et al.Student Model=Qwen2.5-1.5B, K=52026.02 | 99 | 78 | |
| OPTStudent Model=Llama-3.2-3B, K=202026.02 | 99 | 0 | |
| OPTStudent Model=Qwen2.5-1.5B, K=52026.02 | 98 | 0 | |
| He et al.Student Model=Llama-3.2-3B, K=52026.02 | 95 | 80 | |
| He et al.Student Model=Llama-3.1-8B, K=52026.02 | 94 | 76 | |
| KGWStudent Model=Qwen2.5-1.5B, K=202026.02 | 88 | 0 | |
| VIAStudent Model=Qwen2.5-1.5B, K=52026.02 | 84 | 0 | |
| VIAStudent Model=Llama-3.1-8B, K=52026.02 | 77 | 0 | |
| OPTStudent Model=Llama-3.2-3B, K=52026.02 | 55 | 0 | |
| VIAStudent Model=Llama-3.2-3B, K=202026.02 | 52 | 0 | |
| KGWStudent Model=Qwen2.5-1.5B, K=52026.02 | 39 | 0 | |
| VIAStudent Model=Llama-3.2-3B, K=52026.02 | 27 | 0 | |
| GINSEWStudent Model=Qwen2.5-1.5B, K=202026.02 | 8 | 7 | |
| GINSEWStudent Model=Llama-3.1-8B, K=202026.02 | 6 | 16 | |
| GINSEWStudent Model=Llama-3.2-3B, K=202026.02 | 4 | 13 | |
| GINSEWStudent Model=Llama-3.1-8B, K=52026.02 | 1 | 2 | |
| GINSEWStudent Model=Llama-3.2-3B, K=52026.02 | 1 | 1 | |
| GINSEWStudent Model=Qwen2.5-1.5B, K=52026.02 | 1 | 1 | |
| KGWStudent Model=Llama-3.1-8B, K=52026.02 | 0 | 0 | |
| KGWStudent Model=Llama-3.1-8B, K=202026.02 | 0 | 0 | |
| KGWStudent Model=Llama-3.2-3B, K=52026.02 | 0 | 0 | |
| KGWStudent Model=Llama-3.2-3B, K=202026.02 | 0 | 1 |