This project implements a Text-to-SQL engine that converts natural language queries into SQL queries using a GPT-2 model. The engine can be evaluated against the BIRD benchmark to measure its accuracy and performance.
- Natural Language to SQL Conversion: Generates SQL queries from natural language inputs using GPT-2.
- Template-Based SQL Generation: Uses predefined templates for common SQL queries.
- BIRD Benchmark Evaluation: Compares generated SQL queries against a set of benchmark queries from the BIRD dataset.
- Interactive Streamlit Application: Provides an easy-to-use interface for testing queries and evaluating the model.
text_to_sql/
├── bird_benchmark.py # Script to interface with the BIRD benchmark
├── text_to_sql.py # Main Streamlit application
├── requirements.txt # Dependencies
├── bird_questions.json # JSON file containing BIRD benchmark questions
└── README.md # Project documentation
-
Main Application (text_to_sql.py): This is your main Streamlit application script, which contains the logic for generating SQL from natural language queries, executing the SQL, and evaluating it against the BIRD benchmark.
-
BIRD Benchmark Script (bird_benchmark.py): A script dedicated to handling the loading and processing of the BIRD benchmark data. This script could be integrated into text_to_sql.py if desired, but keeping it separate helps in modularity.
-
Dependencies (requirements.txt): This file lists all the Python packages required for your project.
streamlit transformers sqlglot sqlite3 -
BIRD Questions (bird_questions.json): This JSON file contains the questions and expected SQL queries that make up the BIRD benchmark. Ensure this file is in the root directory of your project.
-
Clone the Repository:
git clone https://github.com/your-username/text_to_sql.git cd text_to_sql pip install -r requirements.txt streamlit run text_to_sql
