Skip to content

README.md

Latest commit

 

History

History
131 lines (90 loc) · 3.79 KB

README.md

File metadata and controls

131 lines (90 loc) · 3.79 KB

ZenML Quickstart

Get started with ZenML in minutes. This quickstart demonstrates the core ZenML concepts: pipelines, steps, and artifacts.

🎯 What You'll Learn

  • Create a simple ZenML pipeline
  • Run it locally for development
  • Deploy it as a managed HTTP endpoint
  • Track artifacts and metadata automatically

🚀 Quick Start

Prerequisites

pip install "zenml[server]"
zenml init

Phase 1: Run Locally

Run the pipeline locally to see ZenML in action:

pip install -r requirements.txt
python run.py

This executes a simple pipeline and tracks all artifacts in your local ZenML instance.

Phase 2: View in Dashboard

See your pipeline runs, artifacts, and metadata:

zenml login  # If you haven't already

Open your browser to the ZenML dashboard (URL shown after login) to visualize your pipeline run.

Phase 3: Create a Pipeline Snapshot (Optional)

Create an immutable snapshot of your pipeline for reproducibility and sharing:

zenml pipeline snapshot create pipelines.simple_pipeline.simple_pipeline --name my_snapshot

Snapshots freeze your pipeline code, configuration, and container images so:

  • Reproducibility: Run the exact same pipeline months later
  • Collaboration: Share ready-to-use pipeline configs with teammates
  • Automation: Trigger from dashboards or CI/CD systems without code access

Phase 4: Deploy as an HTTP Service

As well as running a pipeline in batch mode, ZenML allows you to deploy the same pipeline as a real-time HTTP service:

# Deploy your pipeline directly
zenml pipeline deploy pipelines.simple_pipeline.simple_pipeline

# OR deploy a snapshot (if you created one above)
zenml pipeline snapshot deploy my_snapshot --deployment simple_pipeline

Key insight: zenml pipeline deploy automatically creates an implicit snapshot behind the scenes!

Get your deployment URL:

zenml deployment describe simple_pipeline

Phase 5: Invoke Your Deployed Service

Call your deployed pipeline as an HTTP endpoint:

# Using ZenML CLI
zenml deployment invoke simple_pipeline --name="Alice"

# Using curl
ENDPOINT=http://localhost:8000  # Replace with your deployment URL
curl -X POST "$ENDPOINT/invoke" \
  -H "Content-Type: application/json" \
  -d '{"parameters": {"name": "Bob"}}'

Key insight: Your pipeline now runs as a persistent HTTP service. Same code, real-time inference!

Phase 6: Run on Cloud (Optional)

Run the same pipeline on cloud infrastructure without changing any code:

# Create a cloud stack
zenml stack register my_cloud_stack --orchestrator kubeflow --artifact-store s3 --deployer aws...

# Activate it
zenml stack set my_cloud_stack

# Run the same code - it automatically runs on the cloud!
python run.py

# Or deploy the pipeline directly on your cloud infrastructure!
zenml pipeline deploy pipelines.simple_pipeline.simple_pipeline

Key insight: ZenML separates ML code from infrastructure. Your pipeline runs identically locally, deployed as a service, or in the cloud!

📦 Project Structure

quickstart/
├── run.py                      # Main entry point
├── pipelines/
│   └── simple_pipeline.py      # Pipeline definition
├── steps/
│   └── simple_step.py          # Step definitions
├── requirements.txt            # Dependencies
└── configs/
    └── default.yaml            # Pipeline configuration

📚 Next Steps

  • LLM Agent Example: Check out examples/deploying_agent/ for document analysis with an embedded web UI
  • Agent Training Loop: See examples/agent_outer_loop/ for a comprehensive agent with training and evaluation
  • Classical ML: Explore examples/deploying_ml_model/ for traditional ML workflows
  • Documentation: Read the ZenML docs