Vision-language model merging on MM-MergeBench seen tasks
69.23Average Absolute AccuracyIndividual
Evaluation results
| Method | Links | |
|---|---|---|
| IndividualEvaluation Protocol=Seen Tasks (LoRA fine-tuned)2026.03 | 69.23 | |
| Multi TaskEvaluation Protocol=Seen Tasks2026.03 | 63.62 | |
| DC-MergeEvaluation Protocol=Seen Tasks2026.03 | 59.63 | |
| RobustMergeEvaluation Protocol=Seen Tasks2026.03 | 57.33 | |
| Task ArithmeticEvaluation Protocol=Seen Tasks2026.03 | 53.93 | |
| DARE-MergingEvaluation Protocol=Seen Tasks2026.03 | 53.84 | |
| PCB-MergingEvaluation Protocol=Seen Tasks2026.03 | 53.7 | |
| TIES-MergingEvaluation Protocol=Seen Tasks2026.03 | 53.09 | |
| ZeroshotEvaluation Protocol=Seen Tasks2026.03 | 43.37 |