Share your thoughts, 1 month free Claude Pro on usSee more

Abstention Metacognition on AbstentionBench Normal Prompt

67F1 Score

SFT-80K+RL

Updated 1d ago

Evaluation Results

Method	Links
SFT-80K+RL 2026.05		67	89.4	53.6	28
CPT + RL 2026.05		66.8	90.8	52.9	27.2
CPT + RL 2026.05		66.4	92.5	51.8	26.1
SFT-80K+RL 2026.05		66.3	89.8	52.6	27.3
CPT 2026.05		66	91.2	51.6	26.4
SFT (no RL) 2026.05		64.9	90.3	50.7	26.2
CPT + RL 2026.05		64.5	91	49.9	25.6
SFT-80K+RL 2026.05		64.4	92.1	49.8	25.2
Abs-RL 2026.05		64	87.4	50.5	27
DPO+RL 2026.05		63.6	92.4	48.5	24.5
Abs-RL 2026.05		62.5	92	47.2	23.9
Abs-RL 2026.05		62.3	88.5	48.2	25.3
SFT-80K (no RL) 2026.05		62.2	92	46.9	23.7
SFT+RL 2026.05		61.7	88.8	47.3	24.8
DPO (no RL) 2026.05		61.6	90.5	46.7	24.1
SFT+RL 2026.05		60.8	92.8	45.2	22.7
DPO+RL 2026.05		59.2	87.9	44.6	23.7
SFT+RL 2026.05		58.6	91.2	43.2	22.1
SFT-80K+RL 2026.05		57.4	85.6	43.2	23.5
DPO+RL 2026.05		56.6	91.1	41	21
CPT + RL 2026.05		56.5	83.8	42.6	23.7
Abs-RL 2026.05		56	84.6	41.8	23.1
Base 2026.05		54.3	84.5	40.1	22.1
Base 2026.05		54.2	90.6	38.7	19.9
DPO+RL 2026.05		52.6	85.3	38.1	20.8
Base 2026.05		52.1	88.4	36.9	19.4
Base 2026.05		48.3	87.6	33.3	17.7
SFT+RL 2026.05		45.4	86.7	30.8	16.6