List your company
ResearchFormerly WizWand

Knowledge Recovery on TOFU 10% (400 samples) 1.0 (forget)

60ASRREBEL
-2.413.83046.2Feb 5, 2026
Updated 8mo ago

Evaluation results

MethodLinks
REBELUnlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0260
Leak@1000Unlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0245.25
Leak@500Unlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0241.25
REBELUnlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0236.25
Leak@100Unlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0234.5
REBELUnlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0230
REBELUnlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0222
REBELUnlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0222
Leak@1000Unlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0218.75
Leak@10Unlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0218.5
Leak@500Unlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0217.75
REBELUnlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0216.5
Leak@100Unlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0214.75
Leak@1000Unlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0213.25
Leak@500Unlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.0213
Leak@100Unlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.028.75
REBELUnlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.028.25
Leak@10Unlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.027.25
BaselineUnlearning technique=NPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.024.75
Leak@1000Unlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.024.75
Leak@500Unlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.024.5
Leak@10Unlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.023.75
Leak@1000Unlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.023.25
Leak@100Unlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.022.75
Leak@500Unlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.022.75
BaselineUnlearning technique=SimNPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.022
Leak@500Unlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021.75
Leak@1000Unlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021.75
Leak@1000Unlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021.75
BaselineUnlearning technique=AltPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021.5
Leak@500Unlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021.25
Leak@100Unlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.021
Leak@10Unlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.75
Leak@100Unlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.75
BaselineUnlearning technique=IdkNLL, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.5
Leak@10Unlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.5
Leak@100Unlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.5
BaselineUnlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.25
Leak@10Unlearning technique=RMU, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020.25
BaselineUnlearning technique=IdkDPO, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020
BaselineUnlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020
Leak@10Unlearning technique=GradDIFF, Target model=Llama-3.1-1B-Instruct, Hacker model=Qwen2.5-7B-Instruct, Judge model=Qwen2.5-7B-Instruct2026.020