Skip to content

README.md

Latest commit

 

History

History
27 lines (21 loc) · 1.51 KB

README.md

File metadata and controls

27 lines (21 loc) · 1.51 KB

Text-to-SQL with Stable Code 3B

This project explores using Stable Code 3B, a Small Language Model (SLM), for converting natural language questions into SQL queries. It evaluates full fine-tuning and parameter-efficient fine-tuning (LoRA) methods on the BIRD Benchmark dataset.

Project Structure

  • data/: Processed and raw datasets for training and evaluation (generated from BIRD Benchmark).

    • generations/: Model-generated outputs.
    • processed/: JSON files for training, mini-dev, and dev sets (including database schemas).
    • raw/: Original JSON files for BIRD benchmark data (training and dev sets).

  • databases/: SQLite databases (training and dev sets) required for evaluation and schema generation.
    Note: This directory is not included in the repository. Please download the databases from the BIRD Benchmark Dataset webpage.

  • notebooks/: Jupyter notebooks for running full parameter fine-tuning and LoRA fine-tuning.

  • src/: Core project scripts.

    • execution_evaluation.py: Code for evaluating model-generated SQL queries on execution accuracy.
    • schema_linking.py: Handles database schema generation for the training and inference prompts.
    • utils/: Helper scripts for database conversion, execution, and schema linking.

Requirements

Install dependencies from requirements.txt:

pip install -r requirements.txt

Contributions

Aksel Joonas Reedi (akseljoonas), Mika Ernesto Umana Lemus (MikaMann), Mihkel Mariusz Jezierski (mikhelmj)