Question Answering on Pacific unambiguous (dev)
70.95RewardSGP-Oracle
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SGP-Oraclebase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 70.95 | 83.6 | 15.6 | 9.39 | |
| SGPbase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 65.7 | 79.52 | 20.56 | 9.3 | |
| Answerbase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 56.91 | 71.52 | 0 | 0 | |
| Prompted-COTbase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 52.07 | 68.43 | 4.86 | 7.17 | |
| Clarifybase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 46.9 | 68.85 | 100 | 0 | |
| Promptedbase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | 42.44 | 69.51 | 4.82 | 19.15 | |
| Multi-answer (MA)base_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | -4.04 | 63.34 | 0 | 100 | |
| Clarify-and-MAbase_model=Gemma-2-9b, averaging=averaged across rollouts2025.12 | -22.25 | 57.77 | 100 | 100 |