ResearchBenchmarksText Generation on RealToxicityPrompts benignFollow0.555BERTScoreSAD0.530040.536520.5430.54948Apr 28, 2026Evaluation ResultsMethodMethodLinksBERTScoreSADModel=MDLM, Negation S...Model=MDLM, Negation Set=RTP2026.040.555BaseModel=MDLM, Negation S...Model=MDLM, Negation Set=–2026.040.554SADModel=MDLM, Negation S...Model=MDLM, Negation Set=BeaverTails2026.040.553SADModel=MDLM, Negation S...Model=MDLM, Negation Set=ToxiGen2026.040.531