api-reference.md

API Reference

src/fine_tune.py

load_config(path) -> dict

Loads YAML config file.

load_model(cfg) -> (model, tokenizer)

Loads the base model and applies LoRA. All parameters come from config.yaml.

load_and_prepare_dataset(tokenizer, cfg) -> (train_dataset, val_dataset)

Loads the dataset, applies train/validation split, and formats prompts in Alpaca style.

train_model(model, tokenizer, dataset, cfg)

Runs SFT training and saves the model to the configured output directory.

---

src/inference.py

load_model(model_path, load_in_4bit=True) -> (model, tokenizer)

Loads a fine-tuned model for inference.

generate_sql(model, tokenizer, db_schema, question, ...) -> dict

Generates SQL from natural language. Returns {"sql": ..., "explanation": ...}.

parse_response(response) -> dict

Splits raw model output into SQL and explanation parts.

---

src/evaluate.py

load_eval_dataset(dataset_name, num_samples, seed) -> dataset

Loads the saved validation set, or creates a split if not available.

evaluate(model, tokenizer, dataset) -> (results, accuracy)

Runs exact match evaluation and prints progress.

---

CLI Usage

python src/fine_tune.py --config config.yaml
python src/inference.py --hf --schema "..." --question "..."
python src/evaluate.py --hf --num-samples 100 --output results.json