Question Answering on TriviaQA no ctx
59Accuracy (General)Qwen-7B† / Llama-3-8B
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Qwen-7B† / Llama-3-8BGenerator=Qwen-7B†, Verifier=Llama-3-8B2026.03 | 59 | 67 | 8 | 0.823 | 0.794 | 0.752 | 0.634 | 0.993 | 0.953 | 0.976 | 0.564 | |
| Llama-1B / Llama-3-8BGenerator=Llama-1B, Verifier=Llama-3-8B2026.03 | 42 | 67 | 25 | 0.84 | 0.851 | 0.938 | 0.813 | 0.687 | 0.694 | 0.813 | 0.588 | |
| Llama-1B / Qwen-7BGenerator=Llama-1B, Verifier=Qwen-7B2026.03 | 42 | 59 | 17 | 0.84 | 0.851 | 0.738 | 0.806 | 0.655 | 0.675 | 0.758 | 0.506 | |
| Qwen-0.5B / Qwen-7BGenerator=Qwen-0.5B, Verifier=Qwen-7B2026.03 | 21 | 59 | 38 | 0.791 | 0.786 | 0.799 | 0.827 | 0.533 | 0.536 | 0.645 | 0.514 |