Medical Multiple-Choice Question Answering on m23k 750 samples (test)
57.3Average ScoreOurs
Evaluation results
| Method | Links | |||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| OursBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 57.3 | 58.1 | 70.9 | 75.7 | 62.7 | 49.5 | 64.9 | 60.4 | 53.7 | 16.3 | 61.1 | |
| Middle PerplexityBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 55.3 | 56.7 | 69.6 | 74.9 | 62.1 | 45.9 | 59.1 | 57.5 | 48.7 | 16.4 | 61.8 | |
| RandomBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 55.2 | 57.4 | 67.2 | 76.3 | 62.4 | 46 | 62.1 | 54.5 | 49.5 | 15.7 | 61.2 | |
| S2LBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 55.1 | 57.3 | 69.2 | 76 | 61.8 | 48.2 | 63 | 54.2 | 48.1 | 14.6 | 59.1 | |
| Embedding DiversityBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 55 | 57.6 | 67 | 75.7 | 62.8 | 47.1 | 62.3 | 56.7 | 48.2 | 15 | 58.1 | |
| LearnabilityBackbone=Qwen2.5-7B-Instruct, Training samples=750, Mode=Fine-tuning2026.06 | 53.6 | 56.7 | 65.7 | 75.3 | 59.8 | 45 | 59.8 | 52.5 | 46.9 | 14.7 | 59.7 |