Lifelong Unlearning on MLUBench Task A v1.0 (test)
0.97Forget QualityLUMoE
Evaluation results
| Method | Links | ||
|---|---|---|---|
| LUMoEBackbone=LLaVA-7B, Evaluation Stage=Sequential (A-UA, A-UB, A-UD)2026.06 | 0.97 | 0.93 | |
| GABackbone=LLaVA-7B, Evaluation Stage=After Task A (A-UA)2026.06 | 0.375 | 0.22 | |
| GDBackbone=LLaVA-7B, Evaluation Stage=After Task A (A-UA)2026.06 | 0.36 | 0.3 | |
| KLBackbone=LLaVA-7B, Evaluation Stage=After Task A (A-UA)2026.06 | 0.275 | 0.377 | |
| GABackbone=LLaVA-7B, Evaluation Stage=After Task B (A-UB)2026.06 | 0.17 | 0.16 | |
| GABackbone=LLaVA-7B, Evaluation Stage=After Task D (A-UD)2026.06 | 0.135 | 0.115 | |
| GDBackbone=LLaVA-7B, Evaluation Stage=After Task B (A-UB)2026.06 | 0.05 | 0.015 | |
| KLBackbone=LLaVA-7B, Evaluation Stage=After Task B (A-UB)2026.06 | 0.04 | 0.015 | |
| KLBackbone=LLaVA-7B, Evaluation Stage=After Task D (A-UD)2026.06 | 0.04 | 0.007 | |
| GDBackbone=LLaVA-7B, Evaluation Stage=After Task D (A-UD)2026.06 | 0.005 | 0 | |
| NPOBackbone=LLaVA-7B, Evaluation Stage=After Task A, B, or D2026.06 | 0 | 0 |