llama + SQL + PyGWalker + Flask
A personally hosted text-to-sql application that can be used to interact with your databases using natural language queries. The UI has provisions to switch between available databases, search and select tables available in the selected database, modify the SQL query generated from the Llama model and search and visualize the output as needed using PyGWalker.
The application uses code llama 7B GGUF - 5 bit model and can run on CPU (time taken to query the output varies depending on the hardware though).
You can replace this model with other models more suited to your hardware.
Query Pages
Visualization Page
# Clone the repository
git clone https://github.com/raghujhts13/text-to-sql.git
# change directory to where files are present
cd text-to-sql
# install the required libraries using the requirements.txt
pip install -r requirements.txt
# start the application
python app.pychange the user credentials in the .env file inside the config folder FLASK_KEY = random string ( this will not affect the application ) MODEL_PATH = path where you have the saved model (or) where you want to install the model
You can access the application at http://127.0.0.1:5000 (if this is the port where the application is running) If you are unable to do the above, check the command prompt to fetch the URL of the flask server.
You can change the model by altering the variable llm_model inside the app.py file. the first argument is the model name and the second argument is version/ type(second argument is not mandatory, so you can leave it empty if you are unsure about the version) Example I have downloaded the model from TheBloke/CodeLlama-7B-Instruct-GGUF from Hugging Face. Inside this repository I have downloaded the codellama-7b-instruct.Q5_K_M.gguf model. Therefore, I have used the below arguments.
# loading the model using ctransformers
llm_model = llm('TheBloke/CodeLlama-7B-Instruct-GGUF','codellama-7b-instruct.Q5_K_M.gguf')