ML engineering often refers to the process of building and maintaining machine learning models. Thus, it is also similar to MLOps. From my perspective, it is also an standard workflow to build a machine learning model in production.

In my notes, we start from model selection, then model training, and finally model deployment.

Quickstart

  1. Choose a suitable model for your problem
  2. Build a simple interactive web demo

Dive into model training

  1. Train a DETR model with Ray Train and PyTorch-Lightning
  2. Hyperparameter tuning with Ray Tune and Optuna

Let’s make models fast

  1. Accelerate PyTorch Models with ONNX and TensorRT
  2. Serve a model with Ray Serve and NVIDIA Triton

Monitor and debug deployed models

  1. Using Ray Dashboard to monitor and debug models

Engineering Tools

  1. Build a simple web backend with FastAPI
  2. Manage your AI services with docker and docker-compose

Table of contents