List your company

Abstention on AbstentionBench Clean-merged (test)

92.7PrecisionSFT+RL
88.12489.31290.591.688May 30, 2026

Evaluation results

MethodLinks
SFT+RLModel Size=14B, Prompt Type=Normal Prompt2026.0592.751.366
DPO+RLModel Size=14B, Prompt Type=Normal Prompt2026.0592.254.968.8
Abs-RLModel Size=14B, Prompt Type=Normal Prompt2026.0592.153.567.7
CPT+RLModel Size=14B, Prompt Type=Normal Prompt2026.0592.158.671.6
SFT-80K+RLModel Size=14B, Prompt Type=Normal Prompt2026.059256.169.7
SFT+RLModel Size=14B, Prompt Type=Abstention Prompt2026.0591.956.469.9
DPO+RLModel Size=14B, Prompt Type=Abstention Prompt2026.0591.659.271.9
Abs-RLModel Size=14B, Prompt Type=Abstention Prompt2026.0590.261.773.3
SFT-80K+RLModel Size=14B, Prompt Type=Abstention Prompt2026.0588.568.477.2
CPT+RLModel Size=14B, Prompt Type=Abstention Prompt2026.0588.370.178.1