Chart Question Answering on ChartBench
65.11AccuracySketchVL-7B
Evaluation results
| Method | Links | |
|---|---|---|
| SketchVL-7BModel Scale=7B, Training Strategy=FinePO2026.01 | 65.11 | |
| Qwen2.5VL-7BModel Scale=7B2026.01 | 64.78 | |
| SketchVL-3BModel Scale=3B, Training Strategy=FinePO (Full Model)2026.01 | 59.96 | |
| SketchVL-3BModel Scale=3B, Training Strategy=Naive GRPO2026.01 | 59.26 | |
| SketchVL-3BModel Scale=3B, Reward Mechanism=Random scores2026.01 | 58.98 | |
| SketchVL-3BModel Scale=3B, KL Action Regularization=Disabled2026.01 | 58.24 | |
| Qwen2.5VL-3BModel Scale=3B2026.01 | 56.2 | |
| SketchVL-3BModel Scale=3B, Training Strategy=Cold Start only2026.01 | 54.57 | |
| SketchVL-3BModel Scale=3B, Reasoning Method=Zero GRPO (No Sketch)2026.01 | 53.06 | |
| SketchVL-3BModel Scale=3B, Training Strategy=SFT (No RL)2026.01 | 50.24 | |
| VLM-R12026.01 | 39.58 | |
| ChartSketcher-2BModel Scale=2B2026.01 | 30.1 |