Multimodal Performance Retention on Trimmed Average Combined (zero-shot)
24.23ScoreOriginal MLLMs
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Original MLLMsEvaluation Protocol=Zero-shot2025.05 | 24.23 | 51.01 | — | |
| MMEREvaluation Protocol=Merged2025.05 | 24.17 | 50.84 | 99.8 | |
| PCB-MergingEvaluation Protocol=Merged2025.05 | 24.02 | 47.24 | 99.1 | |
| TA (w/ DARE)Evaluation Protocol=Merged2025.05 | 23.95 | 46.5 | 98.8 | |
| TIES (w/ DARE)Evaluation Protocol=Merged2025.05 | 23.84 | 46.37 | 98.4 | |
| TAEvaluation Protocol=Merged2025.05 | 23.82 | 46.23 | 98.3 | |
| TIESEvaluation Protocol=Merged2025.05 | 23.8 | 45.96 | 98.3 | |
| NaiveMC (w/ DARE)Evaluation Protocol=Merged2025.05 | 23.78 | 45.62 | 98.1 | |
| NaiveMCEvaluation Protocol=Merged2025.05 | 23.62 | 44.59 | 97.5 |