Self-confidence estimation on GSM8K easy OOD (test)
89.8AccuracyRLCR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RLCRBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 89.8 | 0.76 | 0.032 | 0.081 | |
| OCBModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 89.7 | 0.847 | 0.039 | 0.073 | |
| OCBBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 89.1 | 0.827 | 0.032 | 0.076 | |
| RLCRModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 88.9 | 0.835 | 0.037 | 0.075 | |
| Group MeanModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 87.8 | 0.849 | 0.03 | 0.079 | |
| Group MeanBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 82.4 | 0.641 | 0.125 | 0.146 |