Multimodal Reasoning on Geo3K (accuracy)
45.59AccuracySFT-Oracle
Evaluation Results
| Method | Links | |
|---|---|---|
| SFT-OracleBackbone=InternVL3-8B2026.05 | 45.59 | |
| VISTA-SFTBackbone=InternVL3-8B2026.05 | 45.59 | |
| R3VBackbone=InternVL3-8B2026.05 | 43.76 | |
| STaR+Backbone=InternVL3-8B2026.05 | 43.43 | |
| CFPOGBackbone=Qwen2.5-VL-7B, Hyperparameters=gamma=0.02 + No Ent2026.06 | 43.18 | |
| STaRBackbone=InternVL3-8B2026.05 | 41.93 | |
| SFT-SeedBackbone=InternVL3-8B2026.05 | 41.6 | |
| GRPOBackbone=Qwen2.5-VL-7B2026.06 | 40.18 | |
| PAPOGBackbone=Qwen2.5-VL-7B2026.06 | 39.98 | |
| SFT-OracleBackbone=Qwen3-VL-2B-Instruct2026.05 | 37.43 | |
| VISTA-SFTBackbone=Qwen3-VL-2B-Instruct2026.05 | 36.27 | |
| R3VBackbone=Qwen3-VL-2B-Instruct2026.05 | 35.77 | |
| STaRBackbone=Qwen3-VL-2B-Instruct2026.05 | 34.28 | |
| STaR+Backbone=Qwen3-VL-2B-Instruct2026.05 | 33.78 | |
| SFT-OracleBackbone=InternVL3-2B2026.05 | 32.78 | |
| SFT-SeedBackbone=Qwen3-VL-2B-Instruct2026.05 | 32.61 | |
| VISTA-SFTBackbone=InternVL3-2B2026.05 | 31.45 | |
| STaR+Backbone=InternVL3-2B2026.05 | 31.28 | |
| STaRBackbone=InternVL3-2B2026.05 | 29.78 | |
| R3VBackbone=InternVL3-2B2026.05 | 29.12 | |
| SFT-SeedBackbone=InternVL3-2B2026.05 | 26.12 |