Text2SQL

A Transformer model trained on WikiSQL dataset that accepts natural language as input and returns SQL Query as output.

Demo

Try it yourself here

Project Organization

---

├── LICENSE
│
├── README.md               <- Documentation to get more information about the project.
│
├── saved_models            <- Trained and serialized models.
│
├── requirements.txt        <- The requirements file for reproducing the environment.
│
├── src                     <- Source code for use in this project.
│   ├── __init__.py         <- Makes src a Python module.
|   |
|   ├── train_tokenizer.py  <- Script to train a sentencepiece tokenizer on the dataset.
|   |
|   ├── dataset.py          <- Script to load and preprocess the dataset.
|   |
|   ├── model.py            <- Script that defines the transformer model.
|   |
|   ├── config.py           <- Contains all the basic parameters for training.
│   │
|   └── train.py            <- Script to train the model. 
│
├── utils
|   ├── examples.csv        <- CSV files with few example predictions.
│   │
|   └── examples.png        <- An image with few example predictions.
|
├── engine.py               <- Script to perform inference on the trained model.
|
└── ui.py                   <- Script to build the streamlit web application.

Running on native machine

dependencies

• python3

pip packages

pip install -r requirements.txt

Steps to train your own model

Scripts

src/train.py - is used to train the model
engine.py - is used to perform inference
ui.py - is used to build the streamlit web application