Kubeflow Trainer
Latest News 🔥
- [2026/08] Kubeflow Trainer v2.3.0 is officially released with the runtime snapshot mechanism for
decoupled runtime lifecycle, enhanced MPI support, and proposal Megatron-LM examples. Check out
the GitHub release notes.
- [2026/03] Kubeflow Trainer v2.2 is officially released with support for JAX and XGBoost
Training Runtimes, enhanced observability with metrics propagation to TrainJob status,
and Flux Framework integration for HPC and MPI workloads. Check out
the blog post announcement.
- [2025/11] Kubeflow Trainer v2.1 is officially released with support of
Distributed Data Cache,
topology aware scheduling with Kueue and Volcano, and LLM post-training enhancements. Check out
the GitHub release notes.
- [2025/09] Kubeflow SDK v0.1 is officially released with support for CustomTrainer,
BuiltinTrainer, and local PyTorch execution. Check out
the GitHub release notes.
- [2025/07] PyTorch on Kubernetes: Kubeflow Trainer Joins the PyTorch Ecosystem. Find the
announcement in the PyTorch blog post.
Overview
Kubeflow Trainer is a Kubernetes-native distributed AI platform for scalable large language model
(LLM) fine-tuning and training of AI models across a wide range of frameworks, including
PyTorch, MLX, HuggingFace, DeepSpeed, Megatron-LM, JAX, XGBoost, and more.
Kubeflow Trainer brings MPI to Kubernetes, orchestrating multi-node, multi-GPU distributed
jobs efficiently across high-performance computing (HPC) clusters. This enables high-throughput
communication between processes, making it ideal for large-scale AI training that requires
ultra-fast synchronization between GPUs nodes.
Kubeflow Trainer seamlessly integrates with the Cloud Native AI ecosystem, including
Kueue for topology-aware scheduling and
multi-cluster job dispatching, Slurm Bridge for
scheduling on hybrid Kubernetes and Slurm clusters, and KAI Scheduler for
GPU aware scheduling.
Kubeflow Trainer reuses existing Kubernetes-native building blocks like
JobSet and LeaderWorkerSet
for AI workload orchestration.
Kubeflow Trainer provides a distributed data cache designed to stream large-scale data with zero-copy
transfer directly to GPU nodes. This ensures memory-efficient training jobs while maximizing
GPU utilization.
With the Kubeflow Python SDK, AI practitioners can effortlessly
develop and fine-tune LLMs while leveraging the Kubeflow Trainer APIs: TrainJob and Runtimes.
Kubeflow Trainer Introduction
Checkout following KubeCon + CloudNativeCon talks for Kubeflow Trainer capabilities:
Additional talks:
Getting Started
Please check the official Kubeflow Trainer documentation
to install and get started with Kubeflow Trainer.
Community
The following links provide information on how to get involved with the Kubeflow Trainer community:
Kubeflow Trainer is a Kubeflow subproject. Refer to
the Kubeflow community page for more information.
Contributing
Please refer to the CONTRIBUTING guide.
Changelog
Please refer to the CHANGELOG directory.
Kubeflow Training Operator V1
Kubeflow Trainer project is currently in alpha status, and APIs may change.
If you are using Kubeflow Training Operator V1, please refer to this migration document.
Kubeflow Community will maintain the Training Operator V1 source code at
the release-1.9 branch.
You can find the documentation for Kubeflow Training Operator V1 in these guides.
Acknowledgement
This project was originally started as a distributed training operator for TensorFlow and later we
merged efforts from other Kubeflow Training Operators to provide a unified and simplified experience
for both users and developers. We are very grateful to all who filed issues or helped resolve them,
asked and answered questions, and were part of inspiring discussions.
We'd also like to thank everyone who's contributed to and maintained the original operators.
