List your company

Priming Attack Robustness on AdvBench No Attack (test)

0ASR (GPT-4o)MOSA
-3.1217.943960.06Oct 1, 2025

Evaluation results

MethodLinks
MOSABase Model=LLaDA2025.10000
RA w/o inter (ablation)Base Model=LLaDA2025.10000
RA (ours)Base Model=LLaDA2025.10000
MOSABase Model=LLaDA 1.52025.10000
RA w/o inter (ablation)Base Model=LLaDA 1.52025.10000
RA (ours)Base Model=LLaDA 1.52025.10000
RA w/o inter (ablation)Base Model=MMaDA2025.10000
RA (ours)Base Model=MMaDA2025.10000
OriginalBase Model=LLaDA2025.10224
OriginalBase Model=LLaDA 1.52025.10226
DPOBase Model=LLaDA2025.1042410
SFTBase Model=LLaDA 1.52025.10666
DPOBase Model=LLaDA 1.52025.106308
MOSABase Model=MMaDA2025.1081212
SFTBase Model=LLaDA2025.101664
DPOBase Model=MMaDA2025.10346457
SFTBase Model=MMaDA2025.10605436
OriginalBase Model=MMaDA2025.10787256