| Dataset Name | SOTA Method | Metric | Trend | ||
|---|---|---|---|---|---|
| WMDP bio | Accuracy71.2 | 93 | 1mo ago | ||
| WMDP cyber | Accuracy47.21 | 47 | 3mo ago | ||
| ZsRE | ROME | Efficiency40.77 | 42 | 2mo ago | |
| MCF | Speed23.35 | 42 | 2mo ago | ||
| WMDP | Performance (Bio)75.9 | 26 | 2mo ago | ||
| RWKU (Forget Set) | FB85.9 | 23 | 3mo ago | ||
| MQuAKE | Efficiency Score64.94 | 21 | 2mo ago | ||
| 16-task Sequential Unlearning Forgotten Data Avg | CORE | CRR90.9 | 18 | 4mo ago | |
| GONE FB Templates Wikidata | Direct Success Rate99.6 | 18 | 4mo ago | ||
| GONE Wikidata (QA Templates) | Direct Success Rate100 | 18 | 4mo ago | ||
| TOFU forget01 LLaDA-8B-Base | random_layer | ForgetLP35.55 | 16 | 1mo ago | |
| 16-task Sequential Unlearning Forgotten Data Last | LwF | Context-aware Refusal Rate (CRR)41.01 | 16 | 4mo ago | |
| FaithUn | FS91.92 | 16 | 4mo ago | ||
| WMDP cyber | BF16 Precision Score41 | 15 | 2mo ago | ||
| WMDP chem | BF16 Score42.3 | 15 | 2mo ago | ||
| WMDP bio | BF16 Score55.7 | 15 | 2mo ago | ||
| The Pile 32 sample (val) | NEO + DPD+ | EL10 (%)0 | 15 | 5mo ago | |
| Internal e-commerce benchmark medium-scale seller 387 items (Forget Set) | ME+GD | ROUGE89.4 | 14 | 5mo ago | |
| MUSE books | Verbosity Score58.4 | 12 | 1mo ago | ||
| WMDP Bio (test) | Refusal Training | Accuracy Forget64.81 | 11 | 5mo ago | |
| MUSE (forget set Df) | VerbMem Df Pre57.9 | 8 | 3mo ago | ||
| MQUAKE Few-shot | Efficiency29.16 | 7 | 2mo ago | ||
| RWKU Utility Set | Fac Score58.2 | 6 | 4mo ago | ||
| RWKU MIA Set | DPO | FM228 | 6 | 4mo ago | |
| RWKU (Neighbor Set) | FB Score95.6 | 6 | 4mo ago |
