Multi-Turn Medical Dialogue on MedQA
68.69AccuracyMEDIQ
Evaluation Results
| Method | Links | |
|---|---|---|
| MEDIQModel=Gemini-2.5-Pro, Method Type=Prompt2026.02 | 68.69 | |
| ATPO (U1 + U2)Model=Qwen3-8B, Method Type=SFT+RL2026.02 | 64.07 | |
| MEDIQModel=GPT-4o, Method Type=Prompt2026.02 | 63.15 | |
| ATPO (U1)Model=Qwen3-8B, Method Type=SFT+RL2026.02 | 62.57 | |
| TreePOModel=Qwen3-8B, Method Type=SFT+RL2026.02 | 61.81 | |
| GRPOModel=Qwen3-8B, Method Type=SFT+RL2026.02 | 57.92 | |
| PPO (H-MDP)Model=Qwen3-8B, Method Type=SFT+RL2026.02 | 57.81 | |
| PPO (MDP)Model=Qwen3-8B, Method Type=SFT+RL2026.02 | 57.38 | |
| ATPO (U1 + U2)Model=Qwen3-4B, Method Type=SFT+RL2026.02 | 55.47 | |
| SFTModel=Qwen3-8B, Method Type=SFT2026.02 | 53.75 | |
| TreePOModel=Qwen3-4B, Method Type=SFT+RL2026.02 | 53.74 | |
| ATPO (U1)Model=Qwen3-4B, Method Type=SFT+RL2026.02 | 53.15 | |
| GRPOModel=Qwen3-4B, Method Type=SFT+RL2026.02 | 51.17 | |
| PPO (MDP)Model=Qwen3-4B, Method Type=SFT+RL2026.02 | 50.6 | |
| DFTModel=Qwen3-8B, Method Type=SFT2026.02 | 48.8 | |
| PPO (H-MDP)Model=Qwen3-4B, Method Type=SFT+RL2026.02 | 48.58 | |
| SFTModel=Qwen3-4B, Method Type=SFT2026.02 | 47.15 | |
| MEDIQModel=Qwen3-8B, Method Type=Prompt2026.02 | 46.03 | |
| DirectModel=Qwen3-8B, Method Type=Prompt2026.02 | 45.22 | |
| DirectModel=Qwen3-4B, Method Type=Prompt2026.02 | 44.94 | |
| ATPO (U1 + U2)Model=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 42.87 | |
| ATPO (U1)Model=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 42.54 | |
| TreePOModel=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 42.05 | |
| DFTModel=Qwen3-4B, Method Type=SFT2026.02 | 41.72 | |
| GRPOModel=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 41.17 | |
| MEDIQModel=Qwen3-4B, Method Type=Prompt2026.02 | 40.11 | |
| PPO (H-MDP)Model=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 39.08 | |
| PPO (MDP)Model=Qwen3-1.7B, Method Type=SFT+RL2026.02 | 38.64 | |
| MEDIQModel=Qwen3-1.7B, Method Type=Prompt2026.02 | 34.2 | |
| DirectModel=Qwen3-1.7B, Method Type=Prompt2026.02 | 34.05 | |
| SFTModel=Qwen3-1.7B, Method Type=SFT2026.02 | 33.42 | |
| DFTModel=Qwen3-1.7B, Method Type=SFT2026.02 | 28.38 |