List your company

Ordinal Regression on Helpsteer

0.72L1 ErrorTask Calibration
0.66241.05121.441.8288May 11, 2026

Evaluation results

MethodLinks
Task CalibrationModel=Qwen3-30B-A3B, Decoding Strategy=Ours2026.050.72
Task CalibrationModel=Gemma-3-4B, Decoding Strategy=Ours2026.050.73
Task CalibrationModel=Qwen-3-4B, Decoding Strategy=Ours2026.050.76
BeamModel=Qwen3-30B-A3B, Decoding Strategy=Beam2026.050.78
MBRModel=Qwen3-30B-A3B, Decoding Strategy=MBR2026.050.78
Task CalibrationModel=Gemma-3-12B, Decoding Strategy=Ours2026.050.79
GreedyModel=Qwen3-30B-A3B, Decoding Strategy=Greedy2026.050.79
AncestralModel=Gemma-3-4B, Decoding Strategy=Ancestral2026.050.8
AncestralModel=Qwen3-30B-A3B, Decoding Strategy=Ancestral2026.050.8
ContrastiveModel=Qwen3-30B-A3B, Decoding Strategy=Contrastive2026.050.81
MBRModel=Gemma-3-4B, Decoding Strategy=MBR2026.050.82
AncestralModel=Qwen-3-4B, Decoding Strategy=Ancestral2026.050.82
BeamModel=Qwen-3-4B, Decoding Strategy=Beam2026.050.82
MBRModel=Qwen-3-4B, Decoding Strategy=MBR2026.050.82
ContrastiveModel=Qwen-3-4B, Decoding Strategy=Contrastive2026.050.83
GreedyModel=Qwen-3-4B, Decoding Strategy=Greedy2026.050.83
BeamModel=Gemma-3-4B, Decoding Strategy=Beam2026.050.85
ContrastiveModel=Gemma-3-4B, Decoding Strategy=Contrastive2026.050.85
GreedyModel=Gemma-3-4B, Decoding Strategy=Greedy2026.050.86
AncestralModel=Gemma-3-12B, Decoding Strategy=Ancestral2026.050.88
ContrastiveModel=Gemma-3-12B, Decoding Strategy=Contrastive2026.050.88
GreedyModel=Gemma-3-12B, Decoding Strategy=Greedy2026.050.89
BeamModel=Gemma-3-12B, Decoding Strategy=Beam2026.050.9
MBRModel=Gemma-3-12B, Decoding Strategy=MBR2026.050.9
Task CalibrationModel=Qwen3-30B-A3B, Decoding Strategy=Ours2026.051.21
Task CalibrationModel=Gemma-3-4B, Decoding Strategy=Ours2026.051.25
Task CalibrationModel=Gemma-3-12B, Decoding Strategy=Ours2026.051.26
Task CalibrationModel=Qwen-3-4B, Decoding Strategy=Ours2026.051.27
MBRModel=Gemma-3-12B, Decoding Strategy=MBR2026.051.89
ContrastiveModel=Gemma-3-12B, Decoding Strategy=Contrastive2026.051.9
GreedyModel=Gemma-3-12B, Decoding Strategy=Greedy2026.051.91
AncestralModel=Gemma-3-12B, Decoding Strategy=Ancestral2026.051.95
BeamModel=Gemma-3-12B, Decoding Strategy=Beam2026.051.95
MBRModel=Gemma-3-4B, Decoding Strategy=MBR2026.052.03
ContrastiveModel=Qwen-3-4B, Decoding Strategy=Contrastive2026.052.03
AncestralModel=Gemma-3-4B, Decoding Strategy=Ancestral2026.052.05
GreedyModel=Gemma-3-4B, Decoding Strategy=Greedy2026.052.05
MBRModel=Qwen-3-4B, Decoding Strategy=MBR2026.052.06
BeamModel=Qwen-3-4B, Decoding Strategy=Beam2026.052.09
BeamModel=Gemma-3-4B, Decoding Strategy=Beam2026.052.1
AncestralModel=Qwen-3-4B, Decoding Strategy=Ancestral2026.052.11
MBRModel=Qwen3-30B-A3B, Decoding Strategy=MBR2026.052.11
ContrastiveModel=Qwen3-30B-A3B, Decoding Strategy=Contrastive2026.052.13
ContrastiveModel=Gemma-3-4B, Decoding Strategy=Contrastive2026.052.14
GreedyModel=Qwen-3-4B, Decoding Strategy=Greedy2026.052.14
GreedyModel=Qwen3-30B-A3B, Decoding Strategy=Greedy2026.052.15
AncestralModel=Qwen3-30B-A3B, Decoding Strategy=Ancestral2026.052.16
BeamModel=Qwen3-30B-A3B, Decoding Strategy=Beam2026.052.16