Is the MLOps tooling the same as the DevOps tooling?
summary
Claims that DevOps principles hold but implementations differ: MLops adds MLflow, Kubeflow, Seldon Core plus feature stores (Feast), experiment trackers, model registries, and extra pipelines for model building. Demonstrates an AWS + Kubernetes landing zone with many microservices and argues reproducibility requires versioning code, data and experiments. Emphasizes monitoring beyond uptime with model drift and data drift, testing challenges from non-deterministic outputs, and that the whole team shares responsibility for running inference servers.
AI-written summary · about
chapters (8)
- 0:00Introduction to MLOps vs DevOps
- 7:27MLOps platform building blocks
- 15:13Key differences in testing
- 20:10Reproducibility challenges
- 22:16Metrics and model monitoring
- 25:48Running inference servers
- 27:56MLOps tooling overview
- 31:11Wrap-up and key takeaways
related talks
14mWhy Your DevOps Skills Are Your MLOps
01mMLOps: Applying DevOps to ML
41mMLOps: Why DevOps Solutions Fall Short in the Machine Learning...- 41mSponsored Session: MLOps: Why DevOps Solutions Fall Short in the Machine Learning World
45mThe Culture, Architecture and Technology of MLOps
30mMLOps for Continuous Learning
01h 13mPractical MLOps- Do we need all the things
41mTHE STATE OF MLOPS - MACHINE LEARNING IN PRODUCTION AT ENTERPRISE SCALE