Multimodal Mathematical Reasoning on Geo3K
50.58AccuracyPSFT
Evaluation Results
| Method | Links | |
|---|---|---|
| PSFTTraining Stage=PSFT, RL Alignment (GRPO)=true2025.08 | 50.58 | |
| BaseTraining Stage=Base, RL Alignment (GRPO)=true2025.08 | 49.42 | |
| SFTTraining Stage=SFT, RL Alignment (GRPO)=true2025.08 | 48.92 | |
| SFTTraining Stage=SFT, RL Alignment (GRPO)=false2025.08 | 33.78 | |
| PSFTTraining Stage=PSFT, RL Alignment (GRPO)=false2025.08 | 33.61 | |
| BaseTraining Stage=Base, RL Alignment (GRPO)=false2025.08 | 29.28 |