Learn experiment tracking with Weights and Biases by following a real text classification training example.
Set up a full MLOps pipeline from model training through Docker packaging by working through the weekly modules.
Understand how to version large data and model files with DVC alongside regular code version control.
Requires a free Weights and Biases account for Week 1 onwards, each week builds on the previous so the series must be followed in order.
MLOps-Basics is a weekly tutorial series that walks through the practical side of taking a machine learning model from an idea to something running reliably in production. MLOps stands for Machine Learning Operations, and it covers all the tasks around a model that are not the model itself: tracking experiments, managing configuration, versioning data, packaging the model, automating tests, and deploying the finished thing. Each week in the series focuses on one topic and uses a concrete text classification project as the running example. The series starts at Week 0 with the basics of loading data, defining a model, and running training using tools from Hugging Face and PyTorch Lightning. From there it adds one layer per week. Week 1 covers experiment tracking with Weights and Biases, a tool that logs metrics and plots as training runs so you can compare different experiments. Week 2 introduces Hydra, a configuration management library that makes it easy to change settings without editing code. Week 3 explains data version control with DVC, a tool for tracking large data and model files the same way Git tracks code. Week 4 goes into ONNX, a format for saving a trained model in a way that lets you run it in different software environments than the one it was trained in. Week 5 covers Docker, the standard tool for packaging an application and its dependencies so it runs the same way anywhere. Week 6 introduces CI/CD with GitHub Actions, which automates running tests and checks whenever new code is pushed. Each week has a companion blog post linked from the README, along with references to documentation and video tutorials. The Jupyter notebooks in the repository contain runnable code for each stage. This is a learning resource, not a production library. Someone who wants to understand what the MLOps discipline involves and how its common tools fit together would work through the weeks in order, reading the blog posts alongside the notebooks.
← graviraja on gitmyhub — every repo by this author, as a profile.
Verify against the repo before relying on details.