Question Answering on MedQA (Q, Ealign)
83.89Q ScoreGRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| GRPOBackbone Model=Ministral 3 8B, Evaluator=GPT-OSS 20B2026.06 | 83.89 | 16.28 | |
| CoRABackbone Model=Ministral 3 8B, Evaluator=GPT-OSS 20B2026.06 | 82.17 | 16.87 | |
| SFTBackbone Model=Ministral 3 8B, Evaluator=GPT-OSS 20B2026.06 | 81.92 | 17.48 | |
| BaseBackbone Model=Ministral 3 8B, Evaluator=GPT-OSS 20B2026.06 | 81.31 | 18.15 | |
| CoRABackbone Model=Gemma 2 9B, Evaluator=GPT-OSS 20B2026.06 | 78.72 | 20.42 | |
| GRPOBackbone Model=Gemma 2 9B, Evaluator=GPT-OSS 20B2026.06 | 72.01 | 26.84 | |
| SFTBackbone Model=Gemma 2 9B, Evaluator=GPT-OSS 20B2026.06 | 71.54 | 28.48 | |
| BaseBackbone Model=Gemma 2 9B, Evaluator=GPT-OSS 20B2026.06 | 69.83 | 29.28 | |
| GRPOBackbone Model=Qwen2.5 7B, Evaluator=GPT-OSS 20B2026.06 | 64.89 | 28.56 | |
| CoRABackbone Model=Qwen2.5 7B, Evaluator=GPT-OSS 20B2026.06 | 63.78 | 32.29 | |
| SFTBackbone Model=Qwen2.5 7B, Evaluator=GPT-OSS 20B2026.06 | 61.26 | 34.79 | |
| BaseBackbone Model=Qwen2.5 7B, Evaluator=GPT-OSS 20B2026.06 | 59.3 | 36.01 |