Unlearning robustness on Forget QA
43.6Accuracy (Original)MMU
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| MMUFine-tuning Scope=Full Model (Vision + LLM)2025.12 | 43.6 | 47.6 | 54.4 | 52 | 53.4 | 53.4 | 9.8 | |
| F+Mask+VKDFine-tuning Scope=Vision2025.12 | 43.2 | 43.6 | 42.8 | 43.2 | 44 | 44 | 0.8 | |
| MANU2025.12 | 31.5 | 35.6 | 33.6 | 36.4 | 36.4 | 37.6 | 6.1 |