Chart Design Prediction on PewResearch (test)
88AccuracyChartDesign
Evaluation Results
| Method | Links | |
|---|---|---|
| ChartDesignbackbone=Phi-3, adaptation=Full FT, training_data=Pew+CharXiV2026.04 | 88 | |
| ChartDesignbackbone=Qwen-3, adaptation=LoRA, training_data=Pew+CharXiV2026.04 | 84 | |
| ChartDesignbackbone=InternVL-2.5, adaptation=LoRA, training_data=Pew+CharXiV2026.04 | 80 | |
| ChartDesignbackbone=Phi-3, adaptation=LoRA, training_data=Pew+CharXiV2026.04 | 78 | |
| ChartDesignbackbone=InternVL-2.5, adaptation=LoRA, training_data=Pew2026.04 | 75 | |
| ChartDesignbackbone=Phi-3, adaptation=LoRA, training_data=Pew2026.04 | 74 | |
| ChartDesignbackbone=Qwen-3, adaptation=LoRA, training_data=Pew2026.04 | 72 | |
| Phi-3adaptation=Zero-shot, parameters=4B, type=backbone baseline2026.04 | 53 | |
| Qwen-3adaptation=Zero-shot, parameters=8B, type=backbone baseline2026.04 | 52 | |
| Qwen-3adaptation=Few-shot prompting (3-shot), parameters=8B2026.04 | 51 | |
| Llama 3.2adaptation=Zero-shot, parameters=11B, type=base model2026.04 | 44 | |
| Llama 3adaptation=Zero-shot, parameters=8B, type=base model2026.04 | 43 | |
| Ministral-3adaptation=Zero-shot, parameters=8B, type=base model2026.04 | 42 | |
| Mistraladaptation=Zero-shot, parameters=7B, type=base model2026.04 | 41 | |
| Gemmaadaptation=Zero-shot, parameters=7B, type=base model2026.04 | 40 | |
| Phi-3adaptation=Zero-shot, parameters=4B, type=base model2026.04 | 38 | |
| InternLM-3adaptation=Zero-shot, parameters=8B, type=base model2026.04 | 35 | |
| DeepSeek-LLMadaptation=Zero-shot, parameters=7B, type=base model2026.04 | 35 | |
| InternLM-2.5adaptation=Zero-shot, parameters=8B, type=base model2026.04 | 32 | |
| InternLM-2.5adaptation=Zero-shot, parameters=8B, type=backbone baseline2026.04 | 32 |