ml-ops

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