List your company

Abstention Metacognition on AbstentionBench Normal Prompt

67F1 ScoreSFT-80K+RL
44.53650.36856.262.032May 30, 2026

Evaluation results

MethodLinks
SFT-80K+RLScale=4B2026.056789.453.628
CPT + RLScale=8B2026.0566.890.852.927.2
CPT + RLScale=14B2026.0566.492.551.826.1
SFT-80K+RLScale=8B2026.0566.389.852.627.3
CPTScale=32B, Backbone=Olmo-3 32B Base, RL=No RL2026.056691.251.626.4
SFT (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0564.990.350.726.2
CPT + RLScale=4B2026.0564.59149.925.6
SFT-80K+RLScale=14B2026.0564.492.149.825.2
Abs-RLScale=8B2026.056487.450.527
DPO+RLScale=14B2026.0563.692.448.524.5
Abs-RLScale=14B2026.0562.59247.223.9
Abs-RLScale=4B2026.0562.388.548.225.3
SFT-80K (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0562.29246.923.7
SFT+RLScale=8B2026.0561.788.847.324.8
DPO (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0561.690.546.724.1
SFT+RLScale=14B2026.0560.892.845.222.7
DPO+RLScale=8B2026.0559.287.944.623.7
SFT+RLScale=4B2026.0558.691.243.222.1
SFT-80K+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0557.485.643.223.5
DPO+RLScale=4B2026.0556.691.14121
CPT + RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0556.583.842.623.7
Abs-RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.055684.641.823.1
BaseScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0554.384.540.122.1
BaseScale=14B2026.0554.290.638.719.9
DPO+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0552.685.338.120.8
BaseScale=4B2026.0552.188.436.919.4
BaseScale=8B2026.0548.387.633.317.7
SFT+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0545.486.730.816.6