README.md

---
base_model:
- griffith-bigdata/Qwen-2.5-Coder-3B-SQL-Writer
---

# FINER-SQL-3B-BIRD

Trained from [`griffith-bigdata/Qwen-2.5-Coder-3B-SQL-Writer`](https://huggingface.co/griffith-bigdata/Qwen-2.5-Coder-3B-SQL-Writer) using GRPO with two dense rewards from the FINER-SQL paper:

🧠 Memory Reward — aligns reasoning with verified traces  
⚙️ Atomic Reward — measures operation-level SQL overlap  

✅ 67.5% EX on BIRD when training only on BIRD train dataset; infer on a single 12-24 GB GPU  

📄 See other models: https://huggingface.co/collections/griffith-bigdata/finer-sql

📄 Github: https://github.com/thanhdath/finer-sql/tree/main


## Citation

```bibtex
@inproceedings{finersql,
  author       = {Thanh Dat Hoang and Thanh Trung Huynh and Matthias Weidlich and Thanh Tam Nguyen and Tong Chen and Hongzhi Yin and Quoc Viet Hung Nguyen},
  title        = {Boosting Small Language Models for Text-to-SQL with Fine-Grained Execution Feedback and Cost-Efficient Rewards},
  booktitle    = {ICDE},
  publisher    = {IEEE},
  year         = {2026},
}
```