Machine Unlearning on ToFU NEK (ROUGE-L, Acc)
0.76ROUGE-LIHL + CE
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| IHL + CETraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.76 | 77 | |
| IHL + CEtraining=adapters only2025.10 | 0.73 | 73 | |
| AStraining=adapters only2025.10 | 0.73 | 76 | |
| ASTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.71 | 73 | |
| IHL + KLTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.69 | 73 | |
| GA + CEtraining=adapters only2025.10 | 0.64 | 68 | |
| IHL + KLtraining=adapters only2025.10 | 0.6 | 60 | |
| GA + CETraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.59 | 57 | |
| ULDTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.57 | 61 | |
| NPO + KLtraining=adapters only2025.10 | 0.55 | 55 | |
| ULDtraining=adapters only2025.10 | 0.55 | 56 | |
| NPO + CEtraining=adapters only2025.10 | 0.4 | 35 | |
| NPO + CETraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.38 | 36 | |
| IHLTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.38 | 45 | |
| NPO + KLTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.37 | 33 | |
| IHLtraining=adapters only2025.10 | 0.32 | 33 | |
| GA + KLtraining=adapters only2025.10 | 0.23 | 24 | |
| NPOtraining=adapters only2025.10 | 0.23 | 26 | |
| NPOTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.21 | 20 | |
| GA + KLTraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.19 | 19 | |
| GAtraining=adapters only2025.10 | 0.17 | 23 | |
| GATraining Protocol=Adapters, Training Data Volume Parity=True2025.10 | 0.12 | 18 |