Multi-turn Calibration on NQ
4.68ECE@1MTCal
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MTCalBackbone=Gemma2-9B-it2026.04 | 4.68 | 2.45 | 4.12 | 22.71 | 4.82 | |
| MTCalBackbone=Llama3.1-8B-Instruct2026.04 | 5.18 | 6.07 | 6.03 | 22.78 | 5.15 | |
| MTCalBackbone=Qwen2.5-7B-Instruct2026.04 | 6.22 | 3.4 | 4.7 | 21.02 | 3.71 | |
| DCalBackbone=Llama3.1-8B-Instruct2026.04 | 6.9 | 11.89 | 7.61 | 23.58 | 7.65 | |
| DCalBackbone=Gemma2-9B-it2026.04 | 10.46 | 11.58 | 14.59 | 23.85 | 13.98 | |
| DCalBackbone=Qwen2.5-7B-Instruct2026.04 | 13.55 | 17.59 | 14.71 | 22.11 | 14.62 | |
| PSBackbone=Gemma2-9B-it2026.04 | 18.39 | 21.55 | 18.93 | 25.22 | 17.15 | |
| SCBackbone=Gemma2-9B-it2026.04 | 20.43 | 21.67 | 28.93 | 32.78 | 22.83 | |
| PSBackbone=Llama3.1-8B-Instruct2026.04 | 20.65 | 33.81 | 25.27 | 29.43 | 24.22 | |
| SCBackbone=Llama3.1-8B-Instruct2026.04 | 20.94 | 21.79 | 22.82 | 28.64 | 17.29 | |
| SLBackbone=Gemma2-9B-it2026.04 | 22.43 | 26.15 | 20.52 | 27.67 | 18.76 | |
| SLBackbone=Llama3.1-8B-Instruct2026.04 | 24.12 | 34.97 | 27.54 | 31.31 | 24.41 | |
| SCBackbone=Qwen2.5-7B-Instruct2026.04 | 25.03 | 28.35 | 26.16 | 30.41 | 19.03 | |
| PSBackbone=Qwen2.5-7B-Instruct2026.04 | 32.02 | 40.09 | 32.65 | 32.92 | 29.63 | |
| P(True)Backbone=Llama3.1-8B-Instruct2026.04 | 32.75 | 43.53 | 41.05 | 40.25 | 34.46 | |
| VerbalBackbone=Qwen2.5-7B-Instruct2026.04 | 33.94 | 45.75 | 43.49 | 43.91 | 39.91 | |
| SLBackbone=Qwen2.5-7B-Instruct2026.04 | 34.11 | 49.51 | 43.96 | 42.05 | 31.93 | |
| P(True)Backbone=Qwen2.5-7B-Instruct2026.04 | 40.54 | 56.3 | 46.46 | 45.22 | 35.83 | |
| VerbalBackbone=Llama3.1-8B-Instruct2026.04 | 41.67 | 51.12 | 43.15 | 41.57 | 36.28 | |
| P(True)Backbone=Gemma2-9B-it2026.04 | 44.37 | 43.63 | 38.21 | 37.5 | 31.22 | |
| VerbalBackbone=Gemma2-9B-it2026.04 | 48.11 | 47.39 | 40.74 | 40.58 | 37.38 |