ZenML

ZenML is an open-source MLOps framework that lets teams write ML and AI pipelines in Python once and run them on any orchestrator and infrastructure stack.

PythonMLMLOpsOpen SourceLLMOps

Visit website GitHub Docs

Stars: 5,610
Forks: 660
Contributors: 144
Latest: 0.97.0

About ZenML

ZenML separates pipeline code from the infrastructure it runs on. Pipelines and steps are plain Python functions decorated with @pipeline and @step; a 'stack' defines where they run (local, Kubernetes, Kubeflow, Airflow, SageMaker, Vertex AI and more) and which artifact store, experiment tracker or model registry to use. ZenML tracks artifacts, lineage and model versions in a central dashboard, so the same pipeline can move from a laptop to production by switching stacks.

Key Features

Pipelines and steps defined with @pipeline and @step decorators

Pluggable stacks: run on local, Kubernetes, Kubeflow, Airflow, SageMaker, Vertex AI and more

Automatic artifact versioning, caching and lineage tracking

Model Control Plane for model versions and promotion

Integrations with MLflow, Weights & Biases, Hugging Face and others

Web dashboard for runs, artifacts and stacks

Pros

Infrastructure-agnostic: switch orchestrators without rewriting pipelines

Strong lineage and artifact tracking out of the box

Large integration catalog across the ML tooling ecosystem

Simple Pythonic API

Cons

Adds an abstraction layer on top of other orchestrators

Some collaboration and RBAC features are limited to ZenML Pro

Frequent releases mean APIs still evolve

Less suited for non-ML data engineering pipelines

Tool Information

Language: Python
License: Apache-2.0
First release: 2020
Latest version: 0.97.0
Maintainer: ZenML GmbH / community

Compare ZenML with other tools

See how ZenML stacks up against other workflow orchestration engines, feature by feature.