Chatbot using a Kaggle Model

Introduction

This small project demonstrates how to build a Chatbot using a Kaggle Models. The model used in Gemma 2 (gemma-2-2b-it-v2).

Model

You will have to download and unzip first the model fromn Kaggle. The model is stored in gemma-2-transformers-gemma-2-2b-it-v2 folder. Alternatively, you can run a short Python script to download the model from your local:

import kagglehub

# Download latest version
path = kagglehub.model_download("google/gemma-2/transformers/gemma-2-2b-it")

print("Path to model files:", path)

When using kagglehub, you will also have to set the Kaggle environment variables.

load_dotenv()
os.environ["KAGGLE_USERNAME"] = os.getenv("kaggle_username")
os.environ["KAGGLE_KEY"] = os.getenv("kaggle_key")

Backend

The backend uses a ModelInterface class and a FastAPI service.

The ModelInterface class initialize the model, loads the tokenizer and the model. It also implements a query function, get_message_response.

In the FastAPI application, the endpoint that receive requests from the frontend, upon user querying the chatbot, is /chat_messages/ [POST].

To start the backend, run:

uvicorn main:app --reload

Frontend

The frontend is implemented using streamlit. When user enters a query, the endpoint from main app is accessed.

To start the frontend, run:

streamlit run app.py

See below an example of usage of the chatbot.