Chart-based Question Answering on EvoChart-QA
58.64AccuracySketchVL-7B
Evaluation results
| Method | Links | |
|---|---|---|
| SketchVL-7BModel Scale=7B, Training Strategy=FinePO2026.01 | 58.64 | |
| Qwen2.5VL-7BModel Scale=7B2026.01 | 54.8 | |
| SketchVL-3BModel Scale=3B, KL Action Regularization=Disabled2026.01 | 48.56 | |
| SketchVL-3BModel Scale=3B, Reward Mechanism=Random scores2026.01 | 48.08 | |
| SketchVL-3BModel Scale=3B, Training Strategy=FinePO (Full Model)2026.01 | 47.28 | |
| SketchVL-3BModel Scale=3B, Training Strategy=Naive GRPO2026.01 | 45.6 | |
| SketchVL-3BModel Scale=3B, Training Strategy=Cold Start only2026.01 | 42.08 | |
| VLM-R12026.01 | 40.32 | |
| Qwen2.5VL-3BModel Scale=3B2026.01 | 39.36 | |
| SketchVL-3BModel Scale=3B, Reasoning Method=Zero GRPO (No Sketch)2026.01 | 30.48 | |
| ChartSketcher-2BModel Scale=2B2026.01 | 26.72 | |
| SketchVL-3BModel Scale=3B, Training Strategy=SFT (No RL)2026.01 | 26.48 |