| license: apache-2.0 | |
| language: | |
| - en | |
| base_model: | |
| - Qwen/Qwen2.5-Coder-1.5B-Instruct | |
| pipeline_tag: translation | |
| ### Performance on the BIRD Development Set | |
| We further evaluate **DatA-SQL-1.5B** on the **BIRD** development set using self-consistency voting. | |
| Under **Vote@8**, our model achieves an **execution accuracy (EX) of 55.86 %**, establishing a new state of the art among open-source Text-to-SQL systems at the 1.5 B scale. | |
| This result demonstrates that even compact models can acquire strong reasoning and SQL generation abilities when trained with our **reasoning-guided data augmentation**. | |
| The lightweight design ensures low training and inference costs while maintaining high execution robustness. | |
| Overall, **DatA-SQL-1.5B** achieves competitive accuracy compared with much larger GPT-based pipelines, highlighting the efficiency and practicality of our approach. | |