Multiple-choice question answering (MCQA) with evidence on RETINA-SAFE (test)
58.91E-AlignQwen2.5-7B
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Qwen2.5-7BEvaluation setting=black-box QA2026.04 | 58.91 | 88.12 | 21.14 | 56.06 | |
| Llama 2-13BEvaluation setting=black-box QA2026.04 | 51.24 | 16.42 | 9.12 | 25.59 | |
| Llama 3-8BEvaluation setting=black-box QA2026.04 | 44.12 | 22.15 | 12.41 | 26.23 | |
| Llama 2-7BEvaluation setting=black-box QA2026.04 | 38.42 | 14.12 | 8.15 | 20.23 | |
| Random-choiceEvaluation setting=4-way; lower bound2026.04 | 25 | 25 | 25 | 25 |