Self-confidence estimation on MATH500 (test)
78.6AccuracyRLCR
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| RLCRBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 78.6 | 0.922 | 0.031 | 0.086 | |
| OCBBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 77.3 | 0.929 | 0.04 | 0.087 | |
| Group MeanBackbone=Qwen2.5-7B-Instruct, Training steps=5122026.02 | 76.6 | 0.893 | 0.032 | 0.098 | |
| OCBModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 75.1 | 0.917 | 0.032 | 0.094 | |
| RLCRModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 75 | 0.914 | 0.043 | 0.093 | |
| Group MeanModel=Qwen2.5-7B-Base, Training steps=5122026.02 | 72.7 | 0.912 | 0.02 | 0.099 |