Hi, I set

configuration.inline.plugins.k8s.gpu-resource-name

to

<http://amd.com/gpu|amd.com/gpu>

in my values when installing flyte-binary. I can confirm that it's set on

/etc/flyte/config.d/100-inline-config.yaml

in the main flyte pod. Yet, when I create a pod with 1 GPU it still sets the resources as

<http://nvidia.com/gpu|nvidia.com/gpu>

. Is there a bug or am I missing something?

Hmm interesting- we have never tried amd- cc @average-finland-92144 / @freezing-boots-56761 do you know

It's probably not AMD-specific. I looked into your code - there is a concept of "resource name" and

<http://nvidia.com/gpu|nvidia.com/gpu>

is the default. I think it's bound to accept arbitrary string. I think it should later appear in the pod's manifest under

limits

. Something goes wrong unfortunately.

hmm i will take a look maybe over the weekend? unless someone knows offhand

What is the purpose of

configuration.inline.plugins.k8s.gpu-resource-name

? If that the setting responsible for accelerator name, then maybe I should just file a bug report?

taking a look @wide-fall-15001

hmm does this also support tpu now?

i guess if you ONLY ever use TPUs in the platform, this will work… but i think if we have gpus resource name override in extended resources, it should handle switching between GPUs and TPUs in a single workflow

Hmm

Sorry, but I'm not even actively using Flyte right now and I didn't figure out how to run development versions from source.

Ohh we can give you a version