Conflict Discrimination on ScienceQA (test)
87.9MCCRPA-SFT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| RPA-SFTbackbone=Qwen3-VL-8B2026.02 | 87.9 | 95.4 | |
| SFTbackbone=Qwen3-VL-8B2026.02 | 87 | 93.1 | |
| RPA-SFTbackbone=InternVL3.5-8B2026.02 | 85.8 | 90.8 | |
| SFTbackbone=InternVL3.5-8B2026.02 | 79.4 | 88.5 | |
| Few-shot CoTbackbone=Qwen3-VL-8B2026.02 | 67.4 | 77.1 | |
| Zero-shotbackbone=Qwen3-VL-8B2026.02 | 64.5 | 74.8 | |
| RPA-SFTbackbone=LLaVA-1.5-7B2026.02 | 58 | 77 | |
| Few-shot CoTbackbone=InternVL3.5-8B2026.02 | 50.8 | 58.5 | |
| Zero-shotbackbone=InternVL3.5-8B2026.02 | 37.3 | 39.2 | |
| SFTbackbone=LLaVA-1.5-7B2026.02 | 34.6 | 72.2 | |
| Zero-shotbackbone=LLaVA-1.5-7B2026.02 | 3.6 | 49.5 | |
| Few-shot CoTbackbone=LLaVA-1.5-7B2026.02 | 3.5 | 48.2 |