Containerized MLflow Image-Caption Model

Pipeline Components

• MLflow tracking
• Deployment of best model via FastAPI
• Streamlit user interface to post data to FastAPI endpoint

Get training data

• Run python sctipt to download training data: python /data/getData.py

Train the model

• MLflow - start MLflow server by running: mlflow ui
• Pytorch training - run "training.py" to train the model

Containerizarion

• User docker compose: docker-compose up -d --build

webUI

• use "http://localhost:8501" to access the service front-end
• upload image and wait for the model to generate the image caption

---

Project Files and Folders

• /backend - Contains files to setup backend service e.g. MLflow and FastAPI
  • /data - Data used for model training
  • /mlruns - ML runs from ML training experiments
  • /mlartifacts - ML artifacts from ML training experiments
  • beheaded_inception3.py - Beheaded Incpection3 model from torchvision
  • Dockerfile - Dockerfile to build backend service
  • main.py - Python script to serve ML artifacts via FastAPI
  • MLproject
  • python_env.yaml
  • requirements-backend.txt - python libraries to be installed during docker image build
  • training.py - Python script PyTorch training with MLflow tracking. Run with this command: python training.py
  • utils.py - utility functions
• /frontend - Folder containining the frontend user interface (UI) aspect of project (i.e. Streamlit)
  • app.py - Python script for the Streamlit web app, connected with FastAPI endpoint for model inference
  • Dockerfile - Dockerfile to build frontend service