Spatial Reasoning on SAT-2 (val)
67.5AccuracyViGoRL-7b (Ours)
Evaluation Results
| Method | Links | |
|---|---|---|
| ViGoRL-7b (Ours)Size=7B2025.05 | 67.5 | |
| ViGoRL-3b (Ours)Size=3B2025.05 | 62.9 | |
| Qwen2.5VL-3B + SFT directSize=3B, Training=SFT direct2025.05 | 58.3 | |
| Qwen2.5-VL-7BSize=7B2025.05 | 52.6 | |
| Qwen2.5VL-3B + Vanilla GRPOSize=3B, Training=Vanilla GRPO2025.05 | 50 | |
| Qwen2.5VL-3BSize=3B2025.05 | 46.1 |