Set up a complete MLflow deployment with experiment tracking, model registry, artifact storage, and CI/CD integration.
You are an MLOps engineer. Set up a production MLflow environment. ## Infrastructure - **Backend Store**: [BACKEND e.g. PostgreSQL / MySQL / SQLite] - **Artifact Store**: [ARTIFACTS e.g. S3 / Azure Blob / GCS / MinIO] - **Deployment**: [DEPLOY e.g. Docker Compose / Kubernetes / AWS ECS] - **Auth**: [AUTH e.g. Basic Auth / OAuth2 / no auth] ## Components ### 1. MLflow Server (`infrastructure/`) - Docker image with MLflow server, [BACKEND] driver, [ARTIFACTS] SDK - Nginx reverse proxy with SSL termination - [AUTH] authentication middleware - Health check endpoint - docker-compose.yml with all services ### 2. Tracking Client Library (`tracking/mlflow_client.py`) - Wrapper class with auto-configuration from environment - Decorators for automatic experiment logging: - `@track_experiment(name, tags)` - Auto-log params, metrics, artifacts - `@log_model(registered_name, stage)` - Auto-register model - Context manager for nested runs (parent/child) - Metric plotting utilities (learning curves, confusion matrices) - Custom artifact loggers for plots, dataframes, config files ### 3. Model Registry Workflows (`registry/`) - Model versioning with automatic semantic versioning - Stage transitions: None → Staging → Production → Archived - Approval workflow with webhook notifications - Model comparison utility: compare metrics across versions - Automated rollback on performance degradation ### 4. CI/CD Integration (`ci/`) - GitHub Actions workflow for model training - Automatic model registration on successful training - Staging deployment on PR creation - Production promotion on PR merge - Model performance gate: fail pipeline if metrics below threshold ### 5. Monitoring Dashboard (`monitoring/`) - Grafana dashboard for experiment metrics - Model performance over time visualization - Resource usage tracking (GPU, memory, training time) - Alert rules for training failures and metric degradation Provide all configuration files, scripts, and a quickstart guide.
Free to copy and use. Compatible with Claude 4 Opus, GPT-5, Gemini 2.0 Flash.
Configure backend store and artifact storage. Deploy MLflow server first, then integrate the tracking client into your training scripts. Set up CI/CD pipelines last.
Initial release
Sign in and download this prompt to leave a review.