Pairwise classification on PMR-Real (Hard)
0.72AccuracyMedGemma-27BSFT
Evaluation Results
| Method | Links | |
|---|---|---|
| MedGemma-27BSFTMethod=UrgentSFT2026.01 | 0.72 | |
| Qwen3-4BSFTMethod=UrgentSFT2026.01 | 0.67 | |
| Qwen3-8BSFTMethod=UrgentSFT2026.01 | 0.64 | |
| Reward-4BurgentMethod=UrgentReward2026.01 | 0.63 | |
| Reward-8BurgentMethod=UrgentReward2026.01 | 0.63 | |
| Reward-4BbaseType=Base2026.01 | 0.61 | |
| Qwen3-8BSize=8B2026.01 | 0.6 | |
| MedGemma-27bSize=27B2026.01 | 0.6 | |
| Qwen3-32BSize=32B2026.01 | 0.58 | |
| Reward-8BbaseType=Base2026.01 | 0.57 | |
| GPT-OSSSize=20B2026.01 | 0.56 | |
| Qwen3-4BSize=4B2026.01 | 0.51 | |
| Qwen3-32B-RSize=32B, Reasoning=True2026.01 | 0.41 |