List your company
ResearchFormerly WizWand

Concept-based Steering on AXBENCH (test)

1.102Overall Steering ScorePrOSV
-0.019120.271940.5630.85406May 7, 2026
Updated 5mo ago

Evaluation results

MethodLinks
PrOSVObjective=SimPO, Model setup=D_Q32B_L322026.051.102
PromptModel setup=D_G9B_L202026.051.075
PromptModel setup=D_Q32B_L322026.051.06
PrOSVObjective=Lang., Model setup=D_Q32B_L322026.051.049
FSSV (RePS) + Joint trainingObjective=SimPO, Model setup=D_Q32B_L322026.050.982
FSSV (RePS)Objective=SimPO, Model setup=D_Q32B_L322026.050.947
FSSV + Joint trainingObjective=Lang., Model setup=D_Q32B_L322026.050.919
PrOSVObjective=SimPO, Model setup=D_G9B_L202026.050.905
FSSV (RePS)Objective=SimPO, Model setup=D_G9B_L202026.050.892
FSSV (RePS) + Joint trainingObjective=SimPO, Model setup=D_G9B_L202026.050.886
PrOSVObjective=Lang., Model setup=D_G9B_L202026.050.859
FSSV + Joint trainingObjective=Lang., Model setup=D_G9B_L202026.050.821
PrOSVObjective=SimPO, Model setup=D_G2B_L102026.050.803
FSSVObjective=Lang., Model setup=D_Q32B_L322026.050.798
FSSVObjective=Lang., Model setup=D_G9B_L202026.050.788
LoReFTModel setup=D_G9B_L202026.050.777
FSSV (RePS) + Joint trainingObjective=SimPO, Model setup=D_G2B_L102026.050.769
PrOSVObjective=Lang., Model setup=D_G2B_L102026.050.758
FSSV (RePS)Objective=SimPO, Model setup=D_G2B_L102026.050.756
FSSV + Joint trainingObjective=Lang., Model setup=D_G2B_L102026.050.736
LoReFTModel setup=D_G2B_L102026.050.701
PromptModel setup=D_G2B_L102026.050.698
FSSVObjective=Lang., Model setup=D_G2B_L102026.050.663
SAEModel setup=D_G9B_L202026.050.546
DiffMeanModel setup=D_G9B_L202026.050.322
DiffMeanModel setup=D_G2B_L102026.050.297
FSSV + Fixed unit factorObjective=Lang., Model setup=D_G2B_L102026.050.072
FSSV + Fixed unit factorObjective=Lang., Model setup=D_G9B_L202026.050.024