neurosql 0.1.1

NeuroSQL: Neural-powered Text-to-SQL generator.

NeuroSQL

NeuroSQL is an advanced neural Text-to-SQL generator based on T5/Transformers.
Convert natural language questions and database table schemas into accurate SQL queries.

---

Installation

NeuroSQL is modular: you can install only the components you need.

Install for production/basic usage (core dependencies only):

pip install neurosql

Install with API support (FastAPI, uvicorn, pydantic):

pip install neurosql[api]

Install for development (includes notebooks, testing, plotting, ML utilities):

pip install neurosql[dev,api]

For exact environment reproduction (e.g., full research replication):

pip install -r requirements.txt

---

Quick Start

Python Example:

from neurosql import NeuroSQLModel

model = NeuroSQLModel.from_pretrained("0xV4h3/neurosql")
sql = model.generate(
    query="Show all sales in Armenia for 2020",
    context="CREATE TABLE sales(id INT, country VARCHAR(20), year INT, sales INT);"
)
print(sql)  # Example: SELECT * FROM sales WHERE country='Armenia' AND year=2020;

---

Command-Line Interface (CLI)

Run SQL generation in the terminal:

neurosql --query "Show sales in Armenia for 2020" --context "CREATE TABLE sales(id INT, country VARCHAR(20), year INT, sales INT);"

---

API

Integrate NeuroSQL into your service using the built-in FastAPI server.
See usage examples in the examples/ directory.

---

Notebooks & Training

Find Jupyter notebooks for fine-tuning and experimenting with the model in the notebooks/ directory.
You can further customize NeuroSQL for your own datasets and domains.

---

Model Card

The pretrained and fine-tuned model is available on Hugging Face:
Hugging Face Model Card

---

Requirements

• Python 3.8+
• torch >= 2.0.0
• transformers >= 4.30.0
• datasets >= 2.10.0
• tqdm
• For API: fastapi, uvicorn, pydantic
• For development: notebook, pytest, matplotlib, scikit-learn, etc.
• See requirements.txt for the complete development environment.

---

License

MIT License