Multiple-Choice Question Answering on SOCIAL SCI. 4-Choice
0.5ΔAccBatch Calibration (BC)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Batch Calibration (BC)Model=Llama3.1-8B-Instruct2026.02 | 0.5 | — | — | -0.006 | |
| RBCorrModel=Llama3.1-8B-Instruct2026.02 | 0.3 | — | — | 0.004 | |
| Batch Calibration (BC)Model=Llama3.1-8B2026.02 | 0 | — | — | 0.037 | |
| RBCorrModel=Llama3.1-8B2026.02 | -0.3 | — | — | 0.04 | |
| Contextual Calibration (CC)Model=Llama3.1-8B-Instruct2026.02 | -1 | — | — | 0.09 | |
| Contextual Calibration (CC)Model=Llama3.1-8B2026.02 | -14 | — | — | 0.348 | |
| BaselineModel=Llama3.1-8B2026.02 | — | 69.7 | 0.038 | — | |
| BaselineModel=Llama3.1-8B-Instruct2026.02 | — | 72.6 | 0.021 | — |