List your company

Abstention Metacognition on AbstentionBench Abstention Prompt

74.9F1 ScoreSFT-80K+RL
55.03660.19365.3570.507May 30, 2026

Evaluation results

MethodLinks
SFT-80K+RLScale=8B2026.0574.985.366.736.5
Abs-RLScale=4B2026.0574.9787243.1
Abs-RLScale=8B2026.0574.383.167.237.7
CPTScale=32B, Backbone=Olmo-3 32B Base, RL=No RL2026.0573.98863.733.7
CPT + RLScale=14B2026.0573.488.962.532.8
SFT-80K+RLScale=4B2026.0573.386.563.634.3
SFT+RLScale=4B2026.0572.584.763.434.9
SFT-80K+RLScale=14B2026.0572.489.16131.9
CPT + RLScale=8B2026.0572.288.560.932.1
DPO+RLScale=4B2026.0572.183.163.735.8
CPT + RLScale=4B2026.0572.186.361.933.5
SFT+RLScale=8B2026.0571.886.661.333
SFT-80K (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0570.889.258.830.7
BaseScale=14B2026.0570.285.659.532.5
DPO+RLScale=8B2026.0569.285.558.131.7
SFT (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0568.889.25629.3
BaseScale=4B2026.0568.679.660.335.3
Abs-RLScale=14B2026.0568.291.154.527.9
DPO (no RL)Scale=32B, Backbone=Olmo-3 32B Base2026.0567.889.654.528.4
DPO+RLScale=14B2026.0566.99252.526.7
Abs-RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0566.377.158.135.2
BaseScale=8B2026.0565.68254.531
SFT+RLScale=14B2026.0564.792.349.925.2
CPT + RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0564.580.853.631
BaseScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.056478.154.232.3
SFT-80K+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0563.580.352.531.2
DPO+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0560.780.548.828.3
SFT+RLScale=3B, Backbone=LLaMA-3.2-3B-Instruct2026.0555.882.84223.7