Large Language Model Watermarking on Mistral-7B-Instruct (test)
1.37Perplexity (PPL)Greedy Decoding
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| Greedy Decoding2024.10 | 1.37 | — | — | — | — | — | — | — | — | |
| Flatk=1, m=10242024.10 | 2.61 | 47.7 | 62.2 | 97.7 | 90 | 94 | 79.7 | 52.8 | 51.1 | |
| Flatk=1, m=5122024.10 | 2.63 | 47.7 | 62.5 | 97.7 | 90 | 94.1 | 79.7 | 54.6 | 51.3 | |
| Aaronson2024.10 | 2.81 | 45.3 | 45.3 | 71.7 | 65.5 | 65.6 | 60.3 | 53.9 | 50.5 | |
| Aaronson Cor.2024.10 | 2.81 | 45.3 | 45.3 | 97.9 | 83.6 | 94.8 | 73.2 | 58.8 | 50.7 | |
| Rec.k=1, m=5122024.10 | 3.05 | 48.3 | 64.5 | 97.2 | 87.9 | 92.6 | 76.5 | 50.4 | 51.2 | |
| Flatk=1, m=322024.10 | 3.06 | 48.4 | 65.3 | 97.8 | 90.2 | 94.2 | 80 | 53 | 50.8 | |
| Flatk=1, m=162024.10 | 3.2 | 47.7 | 64.5 | 97.7 | 89.7 | 93.9 | 79.1 | 52.7 | 51.1 | |
| Rec.k=1, m=322024.10 | 3.29 | 48.4 | 64.3 | 96.3 | 85 | 91.6 | 73.5 | 49.4 | 50.8 | |
| Rec.k=1, m=162024.10 | 3.33 | 49.2 | 66.2 | 95.4 | 82.9 | 90.6 | 71.8 | 53.4 | 50.8 | |
| Flatk=1, m=42024.10 | 3.36 | 50.8 | 67 | 95.8 | 82.9 | 90.3 | 70.5 | 51.3 | 50.6 | |
| Kirchenbauerdelta=12024.10 | 3.37 | 50.1 | 67 | 86.9 | 60.6 | 83.7 | 57.1 | 52.9 | 49.9 | |
| Kirchenbauerdelta=0.52024.10 | 3.39 | 49.6 | 66.6 | 73.2 | 52 | 71 | 51.4 | 49 | 49.8 | |
| Rec.k=1, m=42024.10 | 3.41 | 49 | 65 | 93.4 | 75.5 | 86.3 | 63.2 | 48.4 | 50.4 | |
| Flatk=1, m=22024.10 | 3.46 | 50 | 66.4 | 90.2 | 68.8 | 82 | 58.7 | 50.5 | 50.3 | |
| Random Sampling2024.10 | 3.5 | 49.6 | 65.3 | — | — | — | — | — | — | |
| Kuditipudi2024.10 | 3.55 | 50.3 | 67.3 | 87.8 | 76.6 | 87.2 | 74.4 | 75.9 | 53.2 | |
| Flatk=50, m=322024.10 | 3.67 | 47.3 | 63.9 | 94.2 | 71.6 | 85.5 | 58.1 | 51.1 | 50.5 | |
| Kirchenbauerdelta=22024.10 | 3.69 | 47.9 | 64.1 | 97 | 83.3 | 95.4 | 77.4 | 58.4 | 50.3 | |
| Flatk=50, m=162024.10 | 3.72 | 48.3 | 64.2 | 92.7 | 66.7 | 83.1 | 55.6 | 50.5 | 50.1 | |
| Flatk=50, m=42024.10 | 3.76 | 47.7 | 63.9 | 82.9 | 53.5 | 71.9 | 51.3 | 49.4 | 50 | |
| Rec.k=50, m=162024.10 | 3.77 | 47 | 64 | 81.2 | 54.5 | 73.3 | 51.9 | 51.4 | 50.2 | |
| Flatk=50, m=22024.10 | 3.79 | 48.5 | 64.2 | 69.6 | 50.7 | 62.2 | 50.3 | 47 | 50 | |
| Rec.k=50, m=42024.10 | 3.79 | 48.2 | 63.8 | 74.2 | 51.2 | 65.1 | 50.5 | 46.5 | 49.9 | |
| Rec.k=50, m=322024.10 | 3.79 | 47.2 | 63.3 | 83.3 | 55.7 | 74.4 | 52.2 | 49.4 | 50 | |
| Flatk=10, m=322024.10 | 3.8 | 43 | 60.8 | 98.2 | 89 | 94 | 76.7 | 55 | 50.8 | |
| Flatk=10, m=162024.10 | 3.86 | 44.6 | 60.6 | 97.8 | 87 | 93.1 | 73.5 | 54.3 | 50.7 | |
| Flatk=10, m=42024.10 | 4.06 | 45.2 | 61.5 | 93.8 | 72.7 | 85.7 | 59.4 | 51.3 | 50.3 | |
| Rec.k=10, m=322024.10 | 4.06 | 42.9 | 59.5 | 94.8 | 76.9 | 88.1 | 63.2 | 50.6 | 50.3 | |
| Flatk=10, m=22024.10 | 4.1 | 46.1 | 62.2 | 83.4 | 55.8 | 73.6 | 52 | 49 | 50 | |
| Rec.k=10, m=42024.10 | 4.13 | 45.7 | 61.3 | 88.6 | 61.7 | 78.8 | 53.8 | 48 | 50 | |
| Rec.k=10, m=162024.10 | 4.13 | 43.7 | 59.7 | 93.4 | 74 | 86.8 | 61.6 | 52.9 | 50.4 | |
| Kirchenbauerdelta=32024.10 | 4.67 | 41.5 | 58.4 | 99.3 | 94.4 | 98.6 | 90.9 | 63.4 | 51.5 | |
| Kirchenbauerdelta=42024.10 | 5.81 | 26 | 41.2 | 99.8 | 98.4 | 99.6 | 96.8 | 66.4 | 52.7 |