| name | ml-ops |
| description | Deploy, monitor, and manage ML models in production. Use when setting up model serving, experiment tracking, or ML infrastructure. |
MLOps
Activate this skill when deploying models or managing ML infrastructure.
When to Use
- Deploying models to production endpoints
- Setting up experiment tracking (MLflow, W&B)
- Building model registries and versioning
- Implementing A/B testing for models
- Monitoring model drift and performance
Core Tools
- MLflow: Experiment tracking, model registry
- Weights & Biases: Experiment visualization
- BentoML/Ray Serve: Model serving
- DVC: Data and model versioning
- Evidently AI: Model monitoring
Patterns
import mlflow
mlflow.set_experiment("classification_v2")
with mlflow.start_run():
mlflow.log_params({"lr": 0.001, "epochs": 50})
mlflow.log_metrics({"accuracy": 0.94, "f1": 0.91})
mlflow.sklearn.log_model(model, "model")
Rules
- Version everything: code, data, models, configs
- Automate training pipelines (no manual steps)
- Monitor prediction distributions for drift
- Implement rollback mechanisms for model updates
- Log all experiments, even failed ones