List your company

Calibration on PopQA

0.018ECEmean ablation
-0.004080.144960.2940.44304Apr 1, 2026

Evaluation results

MethodLinks
mean ablationModel=Llama-3.2-3B-Instruct, alpha=–2026.040.0180.16196.967.5
steeringModel=Llama-3.2-3B-Instruct, alpha=0.62026.040.020.16196.567.4
steeringModel=Llama-3.2-3B-Instruct, alpha=0.72026.040.0270.16395.367
steeringModel=Llama-3.2-3B-Instruct, alpha=0.82026.040.0310.16394.667.1
steeringModel=Llama-3.2-3B-Instruct, alpha=0.52026.040.0630.14688.970.5
steeringModel=Qwen2.5-3B-Instruct, alpha=0.52026.040.1110.1377.569.7
steeringModel=Qwen2.5-3B-Instruct, alpha=0.62026.040.1450.14670.766.1
steeringModel=Qwen2.5-3B-Instruct, alpha=0.42026.040.1660.14866.465.7
steeringModel=Qwen2.5-3B-Instruct, alpha=0.72026.040.1670.1666661.4
steeringModel=Qwen2.5-3B-Instruct, alpha=0.82026.040.170.1765.560.5
steeringModel=Llama-3.2-3B-Instruct, alpha=0.42026.040.1960.18965.661.9
steeringModel=Qwen2.5-3B-Instruct, alpha=0.32026.040.2450.19350.355.1
mean ablationModel=Qwen2.5-3B-Instruct, alpha=–2026.040.2950.2444043.1
steeringModel=Llama-3.2-3B-Instruct, alpha=0.32026.040.3380.28740.742
baselineModel=Qwen2.5-3B-Instruct, alpha=–2026.040.4920.43——
baselineModel=Llama-3.2-3B-Instruct, alpha=–2026.040.570.495——