List your company

Concept Steering on AxBench (Held-in)

1.185HMeanFLAS
0.089880.374190.65850.94281May 7, 2026

Evaluation results

MethodLinks
FLASBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 20, T=22026.051.185
HyperSteerBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.051.091
PromptingBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.051.075
FLASBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 20, T=22026.051.025
RePSBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.892
LoReFTBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.777
HyperSteerBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.742
PromptingBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.731
LoReFTBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.722
SFTBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.714
LoRABase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.641
ReFT-r1Base Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.63
RePSBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.606
LoRABase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.602
ReFT-r1Base Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.509
DiffMeanBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.322
AcT (Layer 20)Base Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.27
SAEBase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.191
AcT (All Layers)Base Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.187
SAE-ABase Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.186
DiffMeanBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.178
AcT (All Layers)Base Model=Gemma-2-9B-IT, Intervention Layer=Layer 202026.050.161
SAEBase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.151
AcT (Layer 20)Base Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.144
SAE-ABase Model=Gemma-2-2B-IT, Intervention Layer=Layer 202026.050.132