Task#2 on STAR
94.88AccuracyKGRPO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| KGRPOBackbone=Qwen2.5-VL-7B, Settings=S22025.10 | 94.88 | 97.75 | |
| GPT-4o-miniBackbone=GPT-4o-mini2025.10 | 72.25 | — | |
| S1(Single)Backbone=Qwen2.5-VL-3B2025.10 | 56.63 | 74.75 | |
| GPT-4oBackbone=GPT-4o2025.10 | 56.33 | — | |
| GPT-4vBackbone=GPT-4v2025.10 | 41.25 | — | |
| Zero-shotBackbone=Qwen2.5-VL-3B2025.10 | 20.13 | — |