Abstention on AbstentionBench Clean-merged (test)
92.7PrecisionSFT+RL
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| SFT+RLModel Size=14B, Prompt Type=Normal Prompt2026.05 | 92.7 | 51.3 | 66 | |
| DPO+RLModel Size=14B, Prompt Type=Normal Prompt2026.05 | 92.2 | 54.9 | 68.8 | |
| Abs-RLModel Size=14B, Prompt Type=Normal Prompt2026.05 | 92.1 | 53.5 | 67.7 | |
| CPT+RLModel Size=14B, Prompt Type=Normal Prompt2026.05 | 92.1 | 58.6 | 71.6 | |
| SFT-80K+RLModel Size=14B, Prompt Type=Normal Prompt2026.05 | 92 | 56.1 | 69.7 | |
| SFT+RLModel Size=14B, Prompt Type=Abstention Prompt2026.05 | 91.9 | 56.4 | 69.9 | |
| DPO+RLModel Size=14B, Prompt Type=Abstention Prompt2026.05 | 91.6 | 59.2 | 71.9 | |
| Abs-RLModel Size=14B, Prompt Type=Abstention Prompt2026.05 | 90.2 | 61.7 | 73.3 | |
| SFT-80K+RLModel Size=14B, Prompt Type=Abstention Prompt2026.05 | 88.5 | 68.4 | 77.2 | |
| CPT+RLModel Size=14B, Prompt Type=Abstention Prompt2026.05 | 88.3 | 70.1 | 78.1 |