ResearchBenchmarksDiscriminative Task on AMBERFollow84.3AccuracyMESA70.88474.36777.8581.333Apr 9, 2026Evaluation ResultsMethodMethodLinksAccuracyPrecisionRecallF1 ScoreMESABase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0484.390.385.487.8NulluBase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0481.587.484.285.8VTIBase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0479.182.187.684.8VanillaBase Model=LLaVA-v1.5Base Model=LLaVA-v1.52026.0471.481.973.177.2