Question Answering on GPQA (Accuracy, Consistency, # traces)
2,513# TracesPETS-Offline
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PETS-OfflineBackbone=Qwen-Long, Confidence-weighted=false2026.02 | 2,513 | 100 | 76.3 | |
| PETS-OfflineBackbone=GPT-120B, Confidence-weighted=false2026.02 | 2,580 | 100 | 82.5 | |
| PETS-OfflineBackbone=Qwen3-30B, Confidence-weighted=false2026.02 | 2,607 | 100 | 71.8 | |
| PETS-OfflineBackbone=Qwen3-4B, Confidence-weighted=false2026.02 | 2,780 | 100 | 69.7 | |
| PETS-OfflineBackbone=Qwen-Long, Confidence-weighted=true2026.02 | 2,887 | 99 | 77 | |
| PETS-OfflineBackbone=GPT-20B, Confidence-weighted=false2026.02 | 3,180 | 100 | 75.7 | |
| PETS-OfflineBackbone=GPT-120B, Confidence-weighted=true2026.02 | 3,540 | 99 | 82.1 | |
| PETS-OfflineBackbone=Qwen3-4B, Confidence-weighted=true2026.02 | 3,667 | 99 | 70.3 | |
| PETS-OfflineBackbone=Qwen3-30B, Confidence-weighted=true2026.02 | 3,687 | 99 | 72.1 | |
| PETS-OfflineBackbone=GPT-20B, Confidence-weighted=true2026.02 | 3,693 | 99 | 76.2 | |
| UniformBackbone=Qwen-Long, Confidence-weighted=false2026.02 | 9,973 | 96 | 76.5 | |
| UniformBackbone=Qwen3-30B, Confidence-weighted=false2026.02 | 10,367 | 96 | 71.5 | |
| UniformBackbone=GPT-120B, Confidence-weighted=false2026.02 | 10,393 | 97 | 81.5 | |
| UniformBackbone=GPT-120B, Confidence-weighted=true2026.02 | 10,500 | 96 | 80.8 | |
| UniformBackbone=GPT-20B, Confidence-weighted=false2026.02 | 10,853 | 96 | 75.1 | |
| UniformBackbone=Qwen3-4B, Confidence-weighted=false2026.02 | 11,013 | 96 | 68.9 | |
| UniformBackbone=GPT-20B, Confidence-weighted=true2026.02 | 11,193 | 96 | 75.6 | |
| UniformBackbone=Qwen-Long, Confidence-weighted=true2026.02 | 11,400 | 95 | 76.4 | |
| UniformBackbone=Qwen3-4B, Confidence-weighted=true2026.02 | 11,453 | 95 | 69.9 | |
| UniformBackbone=Qwen3-30B, Confidence-weighted=true2026.02 | 11,607 | 95 | 72.2 |