Pre-training Data Detection on WikiMIA (test)
96AUROCGDS
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GDSTarget Model=LLaMA-7B2026.03 | 96 | 84 | |
| FSDTarget Model=GPT-J-6b2026.03 | 95 | 78 | |
| GDSTarget Model=OPT-6.7b2026.03 | 94 | 67 | |
| GDSTarget Model=GPT-J-6b2026.03 | 93 | 66 | |
| FSDTarget Model=Neo-2.7B2026.03 | 92 | 77 | |
| FSDTarget Model=LLaMA-7B2026.03 | 92 | 41 | |
| GDSTarget Model=Pythia-6.9b2026.03 | 92 | 63 | |
| FSDTarget Model=OPT-6.7b2026.03 | 90 | 63 | |
| FSDTarget Model=Pythia-6.9b2026.03 | 90 | 66 | |
| GDSTarget Model=Neo-2.7B2026.03 | 90 | 60 | |
| Min-k++Target Model=LLaMA-7B2026.03 | 82 | 22 | |
| Min-kTarget Model=LLaMA-7B2026.03 | 73 | 18 | |
| ZLibTarget Model=LLaMA-7B2026.03 | 71 | 22 | |
| Min-k++Target Model=Pythia-6.9b2026.03 | 70 | 18 | |
| PPLTarget Model=LLaMA-7B2026.03 | 69 | 14 | |
| Min-k++Target Model=GPT-J-6b2026.03 | 69 | 19 | |
| Min-kTarget Model=GPT-J-6b2026.03 | 67 | 19 | |
| Min-kTarget Model=Pythia-6.9b2026.03 | 67 | 19 | |
| Min-k++Target Model=Neo-2.7B2026.03 | 67 | 15 | |
| Min-kTarget Model=Neo-2.7B2026.03 | 65 | 17 | |
| Min-k++Target Model=OPT-6.7b2026.03 | 65 | 11 | |
| PPLTarget Model=GPT-J-6b2026.03 | 64 | 13 | |
| PPLTarget Model=Pythia-6.9b2026.03 | 63 | 13 | |
| Min-kTarget Model=OPT-6.7b2026.03 | 63 | 15 | |
| PPLTarget Model=Neo-2.7B2026.03 | 61 | 13 | |
| PPLTarget Model=OPT-6.7b2026.03 | 60 | 12 | |
| ZLibTarget Model=GPT-J-6b2026.03 | 60 | 11 | |
| ZLibTarget Model=Pythia-6.9b2026.03 | 59 | 10 | |
| ZLibTarget Model=Neo-2.7B2026.03 | 58 | 10 | |
| ZLibTarget Model=OPT-6.7b2026.03 | 58 | 10 |