Confidence Calibration on TriviaQA (test)
0.065Expected Calibration ErrorDINCO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DINCOBackbone=Qwen3-32B2025.09 | 0.065 | 0.131 | 0.863 | |
| MSPBackbone=Qwen3-32B2025.09 | 0.113 | 0.164 | 0.792 | |
| SCBackbone=Qwen3-32B2025.09 | 0.129 | 0.19 | 0.741 | |
| VCBackbone=Qwen3-32B2025.09 | 0.153 | 0.158 | 0.862 | |
| SC-VCBackbone=Qwen3-32B2025.09 | 0.166 | 0.213 | 0.707 | |
| K-VCBackbone=Qwen3-32B2025.09 | 0.186 | 0.211 | 0.737 | |
| NVCBackbone=Qwen3-32B2025.09 | 0.194 | 0.169 | 0.88 |