skills.rest
Aidas-dev
Aidas-devCOMMUNITY
Explore all skills by this author →
GitHub

kubeflow-training-operator

Configures legacy Kubeflow Training Operator v1 CRDs for distributed training jobs on Kubernetes.

4|2|Updated Jun 1, 2026
Search Tags:#migration#machine learning#kubernetes#distributed training#kubeflow#mpi#pytorchjob
One-click install
npx skills add https://github.com/Aidas-dev/k8s-agent-skills --skill kubeflow-training-operator-aidas-dev
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: kubeflow-training-operator
Source: https://github.com/Aidas-dev/k8s-agent-skills/tree/main/skills/kubeflow-training-operator
Command: npx skills add https://github.com/Aidas-dev/k8s-agent-skills --skill kubeflow-training-operator-aidas-dev

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams still running the deprecated Kubeflow Training Operator v1 need accurate manifests for PyTorchJob, TFJob, MPIJob, and XGBoostJob, plus clear guidance on migrating these workloads to Trainer v2 TrainJobs. ## Core Features & Use Cases - v1 CRD Manifests: Provides ready-to-adapt YAML examples for PyTorchJob, TFJob, MPIJob, and XGBoostJob with replica specs, restart policies, and GPU resource limits. - Migration Mapping: Maps each v1 CRD to its Trainer v2 equivalent (TrainJob plus Runtime mlPolicy) with per-framework migration status. - Common Pitfalls: Documents shared fields like runPolicy and slotsPerWorker, and warns about missing Kueue integration and gang-scheduling gaps. - Use Case: A platform engineer maintaining a legacy cluster can generate a correct MPIJob manifest with GPU workers, then plan its migration to a TrainJob with an MPI runtime. ## Quick Start Ask the agent to generate a PyTorchJob manifest with one master and three GPU workers, or to explain how to migrate an existing TFJob to Trainer v2.

Frequently Asked Questions about kubeflow-training-operator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I create a PyTorchJob on Kubernetes with Kubeflow?▼

Define a PyTorchJob with apiVersion kubeflow.org/v1 and specify pytorchReplicaSpecs for Master and Worker, each with replicas, restartPolicy, and a pod template. Set nvidia.com/gpu limits in container resources for GPU training.

How do I migrate from Kubeflow Training Operator v1 to Trainer v2?▼

Map each v1 CRD to a TrainJob plus a Runtime with the appropriate mlPolicy: torch for PyTorchJob, mpi for TFJob and MPIJob, and xgboost for XGBoostJob. Follow the official migration guide at kubeflow.org/docs/components/trainer/operator-guides/migration/.

Is Kubeflow Training Operator v1 still maintained?▼

Training Operator v1 is deprecated and maintained only on the release-1.9 branch. New features, performance improvements, and JAX, XGBoost, and Flux support are available only in Trainer v2, so new deployments should use Trainer v2.

Does Kubeflow Training Operator v1 support gang scheduling?▼

v1 jobs have no native Kueue integration. You can use runPolicy.scheduling.podGroupPolicy for basic gang-scheduling, but full queueing and scheduling features require migrating to Trainer v2.

Which frameworks are not yet migrated to Kubeflow Trainer v2?▼

PaddleJob is not yet migrated to Trainer v2 and remains pending. If you rely on PaddlePaddle, pin to specific training-operator v1 versions until Paddle support lands in v2.

Related Skills

Explore More Software Engineering Skills→
NVIDIANVIDIA
3.2k

tao-run-on-kubernetes

Submits TAO container jobs as Kubernetes Jobs with NVIDIA GPU scheduling.

Official
Advanced
sumitakesumitake

mlops-engineer

Designs and hardens model lifecycle infrastructure from training through production serving.

Community
Intermediate
ChanghwanKChanghwanK

devops:storage-ops

Diagnose Rook-Ceph cluster health and PVC issues in the infra-k8s-idc cluster.

Community
Advanced
Z43LZ43L

pytorch-lightning

Organize PyTorch training code into LightningModules with a unified Trainer interface.

Community
Intermediate