Question Answering on TriviaQA (Accg, Accv, AUROC, APGR)
64AccgMistral-7B† / Qwen-7B
Evaluation Results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Mistral-7B† / Qwen-7BGenerator=Mistral-7B†, Verifier=Qwen-7B2026.03 | 64 | 65 | 1 | 84.8 | 30 | 42 | 76.6 | 2.222 | -1.068 | -0.562 | 9.3 | |
| Llama-1B / Qwen-7BGenerator=Llama-1B, Verifier=Qwen-7B2026.03 | 54 | 65 | 10 | 80.5 | 81.1 | 69.1 | 78.8 | 0.696 | 0.724 | 0.764 | 57.4 | |
| Llama-3-8B / Mistral-7BGenerator=Llama-3-8B, Verifier=Mistral-7B2026.03 | 43 | 64 | 20 | 85.8 | 88.7 | 83.3 | 92.3 | 0.818 | 0.838 | 0.858 | 78.5 | |
| Llama-3-8B† / Qwen-7BGenerator=Llama-3-8B†, Verifier=Qwen-7B2026.03 | 43 | 65 | 21 | 85.8 | 88.7 | 77.7 | 87.5 | 0.789 | 0.812 | 0.806 | 75.5 | |
| Qwen-0.5B / Qwen-7BGenerator=Qwen-0.5B, Verifier=Qwen-7B2026.03 | 41 | 65 | 24 | 77.3 | 72.6 | 81.8 | 82.4 | 0.619 | 0.598 | 0.737 | 60.6 |