MLOps Basics
A model is not finished on the day it goes live. The world changes, so the model slowly gets worse. MLOps is the habit of saving versions, watching the model and training it again.
New words
- MLOps
- The work of releasing, watching and updating machine learning models in real use.
- Version
- A saved copy of code, data, a model or a prompt at one point in time.
- Data drift
- The inputs the model sees today are different from the data it learned from.
- Model drift
- The model gets worse over time because the world changed.
- Canary release
- Giving a new version to a small share of users first.
- Monitoring
- Watching live numbers such as quality, speed, cost and errors, with alerts.
- Rollback
- Going back to the last version that worked well.
What you will learn
- The model that forgot the monsoon
- Why models get worse: drift
- Version everything, track every try
- Test first, release to a few, then to all
- Watching, alerts and retraining
- Try it: 12 weeks of a live model
- Trade-offs
- In a real system: LLMOps
- Interview tip
- Key takeaways
10 quiz questions at the end.
The full lesson, with interactive demos and the quiz, opens after you sign in with Google. It is free.