Katib

Overview

Katib is a Kubernetes-native automated machine learning (AutoML) system that manages hyperparameter tuning, neural architecture search, and early stopping. It runs experiments as Kubernetes jobs, leveraging custom resource definitions and a controller to orchestrate trial executions.

Best for

Best for
Teams already using Kubernetes and Kubeflow who need automated hyperparameter tuning

Use cases

• Hyperparameter optimization for models running on Kubernetes
• Neural architecture search integrated with Kubeflow pipelines
• Automated early stopping to reduce wasted compute in training jobs

Notes

Katib is a Kubernetes-native automated machine learning (AutoML) system that manages hyperparameter tuning, neural architecture search, and early stopping. It runs experiments as Kubernetes jobs, leveraging custom resource definitions and a controller to orchestrate trial executions.

1,685 stars on GitHub. Last updated 2026-05-29. Licensed Apache-2.0.

Use cases

• Hyperparameter optimization for models running on Kubernetes
• Neural architecture search integrated with Kubeflow pipelines
• Automated early stopping to reduce wasted compute in training jobs

Pros

• Deep integration with the Kubernetes ecosystem and Kubeflow
• Supports multiple optimization algorithms out of the box
• Scalable to large clusters with parallel trial execution

Cons

• Requires significant Kubernetes expertise to deploy and operate
• Limited to Python-based ML workflows and Kubeflow stack
• Community-driven with slower release cadence than commercial alternatives

Indexed from awesome-llmops and enriched against its public facts.