List your company

Vision-language model merging on MM-MergeBench unseen tasks

39.84Average Absolute AccuracyDC-Merge
24.635228.582632.5336.4774Mar 6, 2026

Evaluation results

MethodLinks
DC-MergeEvaluation Protocol=Unseen Tasks2026.0339.84
RobustMergeEvaluation Protocol=Unseen Tasks2026.0337.99
Multi TaskEvaluation Protocol=Unseen Tasks2026.0336.06
PCB-MergingEvaluation Protocol=Unseen Tasks2026.0333.53
Task ArithmeticEvaluation Protocol=Unseen Tasks2026.0333.31
DARE-MergingEvaluation Protocol=Unseen Tasks2026.0333.15
TIES-MergingEvaluation Protocol=Unseen Tasks2026.0333.14
ZeroshotEvaluation Protocol=Unseen Tasks2026.0325.22