google / 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.

google/xpk Issues

Set cluster versions by default to be based on the recommended rapid channel version
Closed 2 months ago1
Clear up xpk backward compatibility story
Closed 3 months ago1
Create Github runner to execute xpk on vm.
Closed 4 months ago1
Make XPK Handle multiple slice sizes
Updated 5 months ago3
Bump JobSet to v0.3.1 after testing integration
Closed 6 months ago1
Support multiple clusterqueues / localqueues with xpk
Updated 6 months ago
Make sure the user doesn't have the region set using `gcloud config set region` without having a zone set or a zone flag.
Closed 6 months ago1
Add xpk workload label in jobset api
Closed 6 months ago
Wrong output value of TPUVMs when the cluster have the string tpu in its name
Closed 6 months ago1
Create github pip release workflow
Updated 6 months ago
MaxText/XPK story is kind of intertwined
Closed 7 months ago
Graceful Error Message on Failing To Apply Kueue
Closed 7 months ago
Improve CPU Node Pool Defaults
Closed 7 months ago
support positional arguments?
Updated 7 months ago
Set reservation to default capacity type and add Reservation-affinity flag to help users set their reservation type.
Closed 7 months ago
Improve Jobset error messaging to clearly list solution around Kubernetes Permissions
Closed 7 months ago
Make XPK pip3 installable
Closed 7 months ago1
List of Nits From An Early User
Updated 7 months ago
Marrying User Metadata Into Cluster
Updated 7 months ago
Error applying Kueue CRDs on MacOS
Updated 7 months ago1