SVG Drawing on SVG Drawing
51.6DiscoverSFT+DPO
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SFT+DPOModel Backbone=Llama-3.1-8B-Instruct, Recipe=SFT followed by DPO2026.02 | 51.6 | 37 | 44.6 | 2,610 | |
| SFTModel Backbone=Qwen3-8B, Recipe=Supervised Fine-Tuning2026.02 | 48.9 | 38.8 | 70.2 | 2,360 | |
| SFT+DPO+GRPOModel Backbone=Qwen3-8B, Recipe=SFT + DPO + Group Relative Policy Optimization2026.02 | 48.7 | 38.6 | 46.3 | 2,460 | |
| BaseModel Backbone=Qwen3-8B, Strategy=Zero-shot Base2026.02 | 47 | 32 | 54.4 | 3,960 | |
| Prompted BaseModel Backbone=Qwen3-8B, Strategy=Prompted2026.02 | 47 | 35.6 | 75.1 | 2,830 | |
| DPOModel Backbone=Qwen3-8B, Recipe=Direct Preference Optimization2026.02 | 46.9 | 35.1 | 75.9 | 2,700 | |
| BaseModel Backbone=Llama-3.1-8B-Instruct, Strategy=Zero-shot Base2026.02 | 45.6 | 32.5 | 21.6 | 3,590 | |
| SFTModel Backbone=Llama-3.1-8B-Instruct, Recipe=Supervised Fine-Tuning2026.02 | 45.4 | 34.9 | 66.9 | 2,920 | |
| DPOModel Backbone=Llama-3.1-8B-Instruct, Recipe=Direct Preference Optimization2026.02 | 45.3 | 32.5 | 29.2 | 2,890 | |
| COLLABLLMModel Backbone=Llama-3.1-8B-Instruct, Strategy=Collaborative2026.02 | 43 | 29.9 | 30.8 | 3,180 | |
| SFT+DPOModel Backbone=Qwen3-8B, Recipe=SFT followed by DPO2026.02 | 42.6 | 38.7 | 32.7 | 1,810 | |
| Prompted BaseModel Backbone=Llama-3.1-8B-Instruct, Strategy=Prompted2026.02 | 40 | 30.9 | 25.1 | 3,180 |