Uncertainty Quantification on GPQA (test)
55.1AccuracyBase
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| BaseModel=PHI-4, Position=Before2026.05 | 55.1 | 0.312 | 50.2 | 0.25 | |
| UnlearnedModel=PHI-4, Position=Before2026.05 | 55.1 | 0.314 | 48.2 | 0.236 | |
| DropoutModel=PHI-4, Position=Before2026.05 | 55.1 | 0.3 | 57.1 | 0.231 | |
| BaseModel=PHI-4, Position=After2026.05 | 55.1 | 0.396 | 59.1 | 0.399 | |
| UnlearnedModel=PHI-4, Position=After2026.05 | 55.1 | 0.323 | 58.2 | 0.294 | |
| DropoutModel=PHI-4, Position=After2026.05 | 55.1 | 0.292 | 57.4 | 0.237 | |
| BaseModel=GEMMA-12B, Position=Before2026.05 | 32.3 | 0.663 | 46.6 | 0.667 | |
| UnlearnedModel=GEMMA-12B, Position=Before2026.05 | 32.3 | 0.605 | 49.3 | 0.62 | |
| DropoutModel=GEMMA-12B, Position=Before2026.05 | 32.3 | 0.585 | 48.7 | 0.585 | |
| BaseModel=GEMMA-12B, Position=After2026.05 | 32.3 | 0.598 | 48.6 | 0.61 | |
| UnlearnedModel=GEMMA-12B, Position=After2026.05 | 32.3 | 0.33 | 41.7 | 0.324 | |
| DropoutModel=GEMMA-12B, Position=After2026.05 | 32.3 | 0.531 | 52.9 | 0.533 | |
| BaseModel=LLAMA-8B, Position=Before2026.05 | 24.7 | 0.719 | 55 | 0.728 | |
| UnlearnedModel=LLAMA-8B, Position=Before2026.05 | 24.7 | 0.49 | 59.7 | 0.546 | |
| DropoutModel=LLAMA-8B, Position=Before2026.05 | 24.7 | 0.267 | 59.2 | 0.275 | |
| BaseModel=LLAMA-8B, Position=After2026.05 | 24.7 | 0.494 | 68.5 | 0.551 | |
| UnlearnedModel=LLAMA-8B, Position=After2026.05 | 24.7 | 0.32 | 53.2 | 0.306 | |
| DropoutModel=LLAMA-8B, Position=After2026.05 | 24.7 | 0.227 | 62.4 | 0.179 |