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

Abstention Metacognition on AbstentionBench Abstention Prompt

74.9F1 Score

SFT-80K+RL

Updated 1d ago

Evaluation Results

Method	Links
SFT-80K+RL 2026.05		74.9	85.3	66.7	36.5
Abs-RL 2026.05		74.9	78	72	43.1
Abs-RL 2026.05		74.3	83.1	67.2	37.7
CPT 2026.05		73.9	88	63.7	33.7
CPT + RL 2026.05		73.4	88.9	62.5	32.8
SFT-80K+RL 2026.05		73.3	86.5	63.6	34.3
SFT+RL 2026.05		72.5	84.7	63.4	34.9
SFT-80K+RL 2026.05		72.4	89.1	61	31.9
CPT + RL 2026.05		72.2	88.5	60.9	32.1
DPO+RL 2026.05		72.1	83.1	63.7	35.8
CPT + RL 2026.05		72.1	86.3	61.9	33.5
SFT+RL 2026.05		71.8	86.6	61.3	33
SFT-80K (no RL) 2026.05		70.8	89.2	58.8	30.7
Base 2026.05		70.2	85.6	59.5	32.5
DPO+RL 2026.05		69.2	85.5	58.1	31.7
SFT (no RL) 2026.05		68.8	89.2	56	29.3
Base 2026.05		68.6	79.6	60.3	35.3
Abs-RL 2026.05		68.2	91.1	54.5	27.9
DPO (no RL) 2026.05		67.8	89.6	54.5	28.4
DPO+RL 2026.05		66.9	92	52.5	26.7
Abs-RL 2026.05		66.3	77.1	58.1	35.2
Base 2026.05		65.6	82	54.5	31
SFT+RL 2026.05		64.7	92.3	49.9	25.2
CPT + RL 2026.05		64.5	80.8	53.6	31
Base 2026.05		64	78.1	54.2	32.3
SFT-80K+RL 2026.05		63.5	80.3	52.5	31.2
DPO+RL 2026.05		60.7	80.5	48.8	28.3
SFT+RL 2026.05		55.8	82.8	42	23.7