kubeflow/katib

Health v2 · maintenance only

Adoption not part of the score

1697 stars · 536 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

Kubeflow Katib is a Kubernetes-native AutoML system for hyperparameter tuning, early stopping, and neural architecture search. It is ML-framework agnostic and can run training jobs via Kubernetes custom resources such as Kubeflow Training Operator, Argo Workflows, and Tekton Pipelines.

Use cases

• tune hyperparameters of a PyTorch or TensorFlow model on Kubernetes
• run a hyperparameter sweep for an LLM fine-tuning job
• optimize RAG retrieval parameters like top_k and temperature
• perform neural architecture search on a cluster
• apply early stopping to cut wasted training runs
• sweep XGBoost model parameters in parallel
• plug a custom search algorithm into an AutoML pipeline

When to choose

• you already run workloads on Kubernetes and want cluster-native AutoML
• you need framework-agnostic tuning across TensorFlow, PyTorch, XGBoost, or custom code
• you want to scale hyperparameter optimization with distributed trials and early stopping
• you are building a Kubeflow-based ML or GenAI platform

When to avoid

• you have no Kubernetes cluster and just need local hyperparameter search
• you want a simple single-machine tuning library like Optuna
• you need a fully managed AutoML service without infrastructure setup

Facets

service · maturity stable

machine-learning workflow-automation scheduling developer-tools machine-learning artificial-intelligence large-language-models cloud self-hosted python go automl hyperparameter-tuning neural-architecture-search early-stopping kubeflow mlops llm-fine-tuning rag-tuning containers kubernetes docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="kubeflow/katib")