conda create -n sqlft python=3.10 -y
conda activate sqlft
pip install -r requirements.txtRequired for pushing models or pulling gated models:
huggingface-cli loginpython src/fine_tune.pyThis will:
- Load and split the dataset (95% train, 5% validation)
- Fine-tune the model with LoRA
- Save the model to
outputs/sql_translator_model/ - Save the validation set to
outputs/val_dataset/
Edit config.yaml to change hyperparameters, dataset, or model.
python src/inference.py --hf --schema "employees(id, name, salary)" --question "List all employees"
python src/inference.py --hf # interactive modepython src/evaluate.py --hf --num-samples 50For more details, see API Reference.