Closed monirul closed 1 week ago
I spoke with @monirul yesterday about an idea to programmatically verify that feature unification has not taken place. Since we need to do a settings-sdk release for the new models anyways, I think it would be a good idea to make the requisite changes their too.
The basic idea is:
const
booleans for the enabled featureThis was superseded by #4182.
Issue number:
Closes #
Description of changes: This PR introduces new settings API for Nvidia GPUs for Kubernetes Nvidia variants.
New settings are
settings.nvidia-container-runtime.visible-devices-as-volume-mounts
accept-nvidia-visible-devices-as-volume-mounts
value for k8s container-toolkittrue
|false
default:true
settings.nvidia-container-runtime.visible-devices-envvar-when-unprivileged
accept-nvidia-visible-devices-envvar-when-unprivileged
settings of nvidia container runtime for k8s varienttrue
|false
default:false
settings.kubernetes.device-plugins.nvidia.pass-device-specs
pass-device-specs
settings of the device plugin that pass the list of DeviceSpecs to the kubelet on Allocatetrue
|false
default:true
settings.kubernetes.device-plugins.nvidia.device-id-strategy
device-id-strategy
settings of the device plugin which specifies how GPUs are identified and selected for workloads running in a Kubernetes clusteruuid
|index
Default:index
settings.kubernetes.device-plugins.nvidia.device-list-strategy
device-list-strategy
setting in NVIDIA Kubernetes device plugins. It is used to configure how GPUs are listed and allocated to pods in a Kubernetes clusterenvvar
|volume-mounts
default:volume-mounts
Testing done: Yes.
Terms of contribution:
By submitting this pull request, I agree that this contribution is dual-licensed under the terms of both the Apache License, version 2.0, and the MIT license.