https://github.com/ai-hypercomputer/xpk
xpk (Accelerated Processing Kit, pronounced x-p-k,) is a software tool to help Cloud developers to orchestrate training jobs on accelerators such as TPUs and GPUs on GKE.
https://github.com/ai-hypercomputer/xpk
gcloud gke tpu
Last synced: 2 months ago
JSON representation
xpk (Accelerated Processing Kit, pronounced x-p-k,) is a software tool to help Cloud developers to orchestrate training jobs on accelerators such as TPUs and GPUs on GKE.
- Host: GitHub
- URL: https://github.com/ai-hypercomputer/xpk
- Owner: AI-Hypercomputer
- License: apache-2.0
- Created: 2023-06-12T23:14:49.000Z (almost 3 years ago)
- Default Branch: main
- Last Pushed: 2025-02-26T19:51:01.000Z (about 1 year ago)
- Last Synced: 2025-02-26T20:37:11.010Z (about 1 year ago)
- Topics: gcloud, gke, tpu
- Language: Python
- Homepage:
- Size: 2.1 MB
- Stars: 105
- Watchers: 22
- Forks: 31
- Open Issues: 37
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: docs/contributing.md
- License: LICENSE
- Code of conduct: docs/code-of-conduct.md
- Codeowners: .github/CODEOWNERS
Awesome Lists containing this project
README
[](https://github.com/google/xpk/actions/workflows/build_tests.yaml?query=branch%3Amain)
[](https://github.com/google/xpk/actions/workflows/nightly_tests.yaml?query=branch%3Amain)
# Overview
XPK (Accelerated Processing Kit, pronounced x-p-k) is a command line interface that simplifies cluster creation and workload execution on Google Kubernetes Engine (GKE). XPK generates preconfigured, training-optimized clusters and allows easy workload scheduling without any Kubernetes expertise.
XPK is recommended for quick creation of GKE clusters for proofs of concepts and testing.
XPK decouples provisioning capacity from running jobs. There are two structures: clusters (provisioned VMs) and workloads (training jobs). Clusters represent the physical resources you have available. Workloads represent training jobs -- at any time some of these will be completed, others will be running and some will be queued, waiting for cluster resources to become available.
The ideal workflow starts by provisioning the clusters for all of the ML
hardware you have reserved. Then, without re-provisioning, submit jobs as
needed. By eliminating the need for re-provisioning between jobs, using Docker
containers with pre-installed dependencies and cross-ahead of time compilation,
these queued jobs run with minimal start times. Further, because workloads
return the hardware back to the shared pool when they complete, developers can
achieve better use of finite hardware resources. And automated tests can run
overnight while resources tend to be underutilized.
XPK supports a variety of hardware accelerators.
| Accelerator | Type | Recipes |
| :--- | :--- | :--- |
| **Ironwood** | tpu7x | [Run training workload with Ironwood and regular/gSC/DWS Calendar reservations using GCS Bucket storage](./docs/usage/tpu7x/recipes/reservation_gcs_bucket_recipe.md)
[Run training workload with Ironwood with flex-start using Filestore storage](./docs/usage/tpu7x/recipes/flex_filestore_recipe.md)
[Run training workload with Ironwood and flex-start using Lustre storage](./docs/usage/tpu7x/recipes/flex_lustre_recipe.md) |
| **Trillium** | v6e | [Create Cluster](./docs/usage/clusters.md)
[Create Workload](./docs/usage/workloads.md) |
| **TPU v5p** | v5p | [Create Cluster](./docs/usage/clusters.md)
[Create Workload](./docs/usage/workloads.md) |
| **TPU v5e** | v5e | [Create Cluster](./docs/usage/clusters.md)
[Create Workload](./docs/usage/workloads.md) |
| **TPU v4** | v4 | [Create Cluster](./docs/usage/clusters.md)
[Create Workload](./docs/usage/workloads.md) |
| **GPU A4X** | gb200 | [Create Cluster](./docs/usage/gpu.md)
[Create Workload](./docs/usage/workloads.md) |
| **GPU A4** | b200 | [Create Cluster](./docs/usage/clusters.md#provisioning-a3-ultra-a3-mega-and-a4-clusters-gpu-machines)
[Create Workload](./docs/usage/workloads.md#workloads-for-a3-ultra-a3-mega-and-a4-clusters-gpu-machines) |
| **GPU A3 Ultra** | h200 | [Create Cluster](./docs/usage/clusters.md#provisioning-a3-ultra-a3-mega-and-a4-clusters-gpu-machines)
[Create Workload](./docs/usage/workloads.md#workloads-for-a3-ultra-a3-mega-and-a4-clusters-gpu-machines) |
| **GPU A3 Mega** | h100-mega | [Create Cluster](./docs/usage/clusters.md#provisioning-a3-ultra-a3-mega-and-a4-clusters-gpu-machines)
[Create Workload](./docs/usage/workloads.md#workloads-for-a3-ultra-a3-mega-and-a4-clusters-gpu-machines) |
| **GPU A3 High** | h100 | [Create Cluster](./docs/usage/gpu.md)
[Create Workload](./docs/usage/workloads.md) |
| **GPU A100** | A100 | [Create Cluster](./docs/usage/gpu.md)
[Create Workload](./docs/usage/workloads.md) |
| **CPU** | n2-standard-32 | [Create Cluster](./docs/usage/cpu.md)
[Create Workload](./docs/usage/workloads.md) |
XPK also supports the following [Google Cloud Storage solutions](./docs/usage/storage.md):
| Storage Type | Documentation |
| ------------------------------------------ | ----------------------------------------------------------------------- |
| Cloud Storage FUSE | [docs](./docs/usage/storage.md#fuse) |
| Filestore | [docs](./docs/usage/storage.md#filestore) |
| Parallelstore | [docs](./docs/usage/storage.md#parallelstore) |
| Block storage (Persistent Disk, Hyperdisk) | [docs](./docs/usage/storage.md#block-storage-persistent-disk-hyperdisk) |
# Documentation
- [Permissions](./docs/permissions.md)
- [Installation](./docs/installation.md)
- Usage:
- [Clusters](./docs/usage/clusters.md)
- [GPU](./docs/usage/gpu.md)
- [CPU](./docs/usage/cpu.md)
- [Autoprovisioning](./docs/usage/autoprovisioning.md)
- [Workloads](./docs/usage/workloads.md)
- [Docker](./docs/usage/docker.md)
- [Storage](./docs/usage/storage.md)
- [Advanced](./docs/usage/advanced.md)
- [Inspector](./docs/usage/inspector.md)
- [Troubleshooting](./docs/troubleshooting.md)
# Dependencies
| Dependency | When used |
| ------------------------------------------------------------------------------------------------------------ | --------------------------- |
| [Google Cloud SDK (gcloud)](https://cloud.google.com/sdk/docs/install) | _always_ |
| [kubectl](https://cloud.google.com/kubernetes-engine/docs/how-to/cluster-access-for-kubectl#install_kubectl) | _always_ |
| [ClusterToolkit](https://github.com/GoogleCloudPlatform/cluster-toolkit) | Provisioning GPU clusters |
| [Kueue](https://github.com/kubernetes-sigs/kueue) | Scheduling workloads |
| [JobSet](https://github.com/kubernetes-sigs/jobset) | Workload creation |
| [Docker](https://docs.docker.com/engine/install/) | Building workload container |
| [CoreDNS](https://github.com/coredns/deployment/tree/master/kubernetes) | Cluster set up |
| [PathwaysJob](https://github.com/google/pathways-job) | Running Pathways workloads |
# Privacy notice
To help improve XPK, feature usage statistics are collected and sent to Google. You can opt-out at any time by executing
the following shell command:
```shell
xpk config set send-telemetry
```
XPK telemetry overall is handled in accordance with the [Google Privacy Policy](https://policies.google.com/privacy). When
you use XPK to interact with or utilize GCP Services, your information is handled in accordance with the
[Google Cloud Privacy Notice](https://cloud.google.com/terms/cloud-privacy-notice).
# Contributing
Please read [`contributing.md`](./docs/contributing.md) for details on our code of conduct, and the process for submitting pull requests to us.
# Get involved
We'd love to hear from you! If you have questions or want to discuss ideas, join us on [GitHub Discussions](https://github.com/AI-Hypercomputer/xpk/discussions). Found a bug or have a feature request? Please let us know on [GitHub Issues](https://github.com/AI-Hypercomputer/xpk/issues).
# License
This project is licensed under the Apache License 2.0 - see the [`LICENSE`](./LICENSE) file for details