Next Action Prediction on MedVidBench
44.2NAP AccuracyQwen2.5VL-7B_SFT
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5VL-7B_SFTTraining=SFT baseline2025.12 | 44.2 | |
| Qwen2.5VL-7B_MedGRPOTraining=MedGRPO2025.12 | 42.7 | |
| GPT-4.1Type=off-the-shelf2025.12 | 25 | |
| Gemini-2.5-flashType=off-the-shelf2025.12 | 22.8 | |
| Qwen2.5VL-7BType=off-the-shelf2025.12 | 15.1 |