List your company

Over-refusal Evaluation on NQ (Natural Questions)

0ORRVectorSteer
-1.69.22030.8Jan 27, 2026

Evaluation results

MethodLinks
VectorSteerModel=Mistral-v0.3, Size=7B2026.010
OriginalModel=Phi-4, Size=15B2026.010
LLM-VAModel=Phi-4, Size=15B2026.010
VectorSteerModel=Qwen2.5, Size=14B2026.010
OriginalModel=Qwen3, Size=4B2026.010
AlphaSteerModel=Qwen3, Size=4B2026.010
LLM-VAModel=Qwen3, Size=4B2026.010
AlphaSteerModel=Qwen3, Size=8B2026.010
LLM-VAModel=Qwen3, Size=8B2026.010
OriginalModel=Qwen3, Size=14B2026.010
AlphaSteer+Model=Qwen3, Size=14B2026.010
AlphaSteerModel=Qwen3, Size=14B2026.010
VectorSteerModel=Qwen3, Size=14B2026.010
LLM-VAModel=Qwen3, Size=14B2026.010
AlphaSteer+Model=Mistral-v0.3, Size=7B2026.012
AlphaSteerModel=Mistral-v0.3, Size=7B2026.012
AlphaSteer+Model=Phi-4, Size=15B2026.012
AlphaSteerModel=Phi-4, Size=15B2026.012
AlphaSteer+Model=Qwen2.5, Size=7B2026.012
AlphaSteerModel=Qwen2.5, Size=7B2026.012
AlphaSteer+Model=Qwen2.5, Size=14B2026.012
AlphaSteerModel=Qwen2.5, Size=14B2026.012
LLM-VAModel=Qwen2.5, Size=14B2026.012
AlphaSteer+Model=Qwen3, Size=4B2026.012
VectorSteerModel=Qwen3, Size=4B2026.012
SCANSModel=Qwen3, Size=4B2026.012
OriginalModel=Qwen3, Size=8B2026.012
VectorSteerModel=Qwen3, Size=8B2026.012
AlphaSteer+Model=gemma-2, Size=9B2026.014
AlphaSteerModel=gemma-2, Size=9B2026.014
OriginalModel=Mistral-v0.3, Size=7B2026.014
SCANSModel=Mistral-v0.3, Size=7B2026.014
LLM-VAModel=Phi-3.5, Size=4B2026.014
VectorSteerModel=Phi-4, Size=15B2026.014
OriginalModel=Qwen2.5, Size=7B2026.014
SCANSModel=Qwen2.5, Size=7B2026.014
LLM-VAModel=Qwen2.5, Size=7B2026.014
OriginalModel=Qwen2.5, Size=14B2026.014
OriginalModel=Llama-3.1, Size=8B2026.016
AlphaSteerModel=Llama-3.1, Size=8B2026.016
LLM-VAModel=gemma-2, Size=9B2026.016
OriginalModel=Phi-3.5, Size=4B2026.016
AlphaSteer+Model=Phi-4, Size=4B2026.016
AlphaSteerModel=Phi-4, Size=4B2026.016
VectorSteerModel=Phi-4, Size=4B2026.016
SCANSModel=Phi-4, Size=15B2026.016
SCANSModel=Qwen2.5, Size=3B2026.016
LLM-VAModel=Phi-4, Size=4B2026.018
SCANSModel=Qwen2.5, Size=14B2026.018
AlphaSteer+Model=Llama-3.1, Size=8B2026.0110
LLM-VAModel=Llama-3.1, Size=8B2026.0110
AlphaSteerModel=Phi-3.5, Size=4B2026.0110
AlphaSteer+Model=Qwen2.5, Size=3B2026.0110
AlphaSteerModel=Qwen2.5, Size=3B2026.0110
AlphaSteer+Model=Qwen3, Size=8B2026.0110
SCANSModel=Qwen3, Size=8B2026.0110
VectorSteerModel=Llama-3.1, Size=8B2026.0112
SCANSModel=gemma-2, Size=9B2026.0112
LLM-VAModel=Mistral-v0.3, Size=7B2026.0112
AlphaSteer+Model=Phi-3.5, Size=4B2026.0112
SCANSModel=Llama-3.1, Size=8B2026.0114
OriginalModel=Qwen2.5, Size=3B2026.0114
OriginalModel=gemma-2, Size=9B2026.0116
OriginalModel=Phi-4, Size=4B2026.0116
LLM-VAModel=Qwen2.5, Size=3B2026.0116
VectorSteerModel=gemma-2, Size=9B2026.0118
VectorSteerModel=Phi-3.5, Size=4B2026.0118
VectorSteerModel=Qwen2.5, Size=7B2026.0124
VectorSteerModel=Qwen2.5, Size=3B2026.0132
SCANSModel=Phi-4, Size=4B2026.0136
SCANSModel=Phi-3.5, Size=4B2026.0140
SCANSModel=Qwen3, Size=14B2026.0140