Set up and manage Cloud TPU resources
-
Set up a project and enable the Cloud TPU API
Set up a Google Cloud project and enable the Cloud TPU API.
-
Connect to Cloud Storage buckets
Store your machine learning data on a Cloud Storage bucket. Give your Cloud TPU access to data objects in the bucket.
-
Manage TPUs
Manage Cloud TPU resources.
-
Manage Queued Resources
Request Cloud TPU in a queued manner.
-
Monitor Cloud TPU VMs
Monitor Cloud TPU VMs.
-
Monitor Cloud TPU Nodes
Monitor Cloud TPU Nodes.
-
Manage maintenance events with Cloud TPU Pods
Understand Compute Engine VM maintenance events and how to recover a Cloud TPU Pod after a maintenance event.
-
Scale ML workloads using Ray
Use the Cloud TPU Ray tool to scale ML workloads.
-
Preemptible TPUs
How and why you can allow Cloud TPU to preempt your TPU for usage by other workloads.
-
Cloud TPU audit logs
Access and use Cloud TPU audit logs.
-
Switch software versions on your Cloud TPU
Switch software versions on your Cloud TPU.
-
Add a persistent disk to a TPU VM
Add a persistent disk to a TPU VM to expand your local disk capacity.
Prepare datasets
-
Download, pre-process, and upload the ImageNet dataset
How to download, preprocess, and upload the ImageNet dataset to a Cloud Storage bucket.
-
Download, pre-process, and upload the COCO dataset
How to download, preprocess, and upload the COCO dataset to a Cloud Storage bucket.
-
Convert an image classification dataset for use with Cloud TPU
Use the data converter sample script to convert an image classification data set into the TFRecord format used to train Cloud TPU models.
-
Advanced guide to Inception v3
An advanced guide to running Inception v3 on Cloud TPU.
Cloud TPU performance guides
-
Profile your model on Cloud TPU Nodes
Monitoring and profiling Cloud TPU Nodes using TensorBoard and command-line tools.
-
Profile your model on Cloud TPU VMs
Monitoring and profiling Cloud TPU VMs using TensorBoard and command-line tools.
-
Cloud TPU performance guide
Troubleshoot Cloud TPU performance issues.
-
TensorFlow performance profiling
Walks you through how to use Cloud TPU performance tools and the metrics auto-analysis feature with TensorFlow.
-
PyTorch XLA performance profiling
Walks you through how to use Cloud TPU performance tools and the metrics auto-analysis feature with PyTorch.
-
JAX performance profiling guide
An introduction to running JAX code on TPU Pod slices.
Containers
-
Run Cloud TPU applications on GKE
A guide to setting up Cloud TPU with Google Kubernetes Engine.
-
GKE Cluster with Cloud TPU using a Shared VPC
How to set up a GKE Cluster with Cloud TPU configuration managed by a Shared VPC network.
-
Run Cloud TPU applications in a Docker container
Run Cloud TPU applications in a Docker container.