Text-to-JSON on DeepJSONEval
93.38Format AdherenceQwen3-4B SFT + STAGE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3-4B SFT + STAGEBase Model=Qwen3-4B SFT, Training Data=STAGE2026.06 | 93.38 | 84.68 | 62.17 | |
| Qwen2.5-3B + STAGEBase Model=Qwen2.5-3B, Training Data=STAGE2026.06 | 92.68 | 80.16 | 41.62 | |
| Qwen3-4BBase Model=Qwen3-4B2026.06 | 91.44 | 83.37 | 51.91 | |
| Qwen3-4B + JSONSchemaBenchBase Model=Qwen3-4B, Training Data=JSONSchemaBench2026.06 | 88.03 | 77.78 | 42.67 | |
| Qwen3-4B-Thinking-2507Base Model=Qwen3-4B-Thinking-25072026.06 | 79.97 | 73.3 | 47.13 | |
| Qwen3-4B + ScrapeGraphAIBase Model=Qwen3-4B, Training Data=ScrapeGraphAI2026.06 | 78.43 | 70.99 | 44.38 | |
| Qwen3-4B + GlaiveBase Model=Qwen3-4B, Training Data=Glaive2026.06 | 76.61 | 70.28 | 46.89 | |
| Llama-3.2-3B-Instruct + STAGEBase Model=Llama-3.2-3B-Instruct, Training Data=STAGE2026.06 | 61.56 | 48.34 | 17.51 | |
| Qwen2.5-3BBase Model=Qwen2.5-3B2026.06 | 47.52 | 41.31 | 21.54 | |
| Llama-3.2-1B-Instruct + STAGEBase Model=Llama-3.2-1B-Instruct, Training Data=STAGE2026.06 | 38.03 | 21.04 | 2.76 | |
| Llama-3.2-3B-InstructBase Model=Llama-3.2-3B-Instruct2026.06 | 22.45 | 18.77 | 9.61 | |
| Llama-3.2-1B-InstructBase Model=Llama-3.2-1B-Instruct2026.06 | 2.25 | 1.41 | 0.22 |