Does gpu parameter in

@resource

decorater works well? When using with

--with kubernetes:image=my_image

in cli run, the job in k8s requests cpu&memory as written in resource decorator but it doesn't request gpu. I wanna check if this problem happens only in my setting or not.

how have you configured the gpu machines in the kubernetes cluster?

yes it works well if I write all resources in cli parameter. e.g.) --with kubernetes:image=my_image,gpu=1,cpu=2,memory=3

and it doesn't if you use

@kubernetes

?

With

@kubernetes

decorator, it also works well. With

@resource

decorator +

--with kubernetes

, no it doesn't/

that may be a regression with Metaflow

Yes I guess so if it's reproducible for others. I thought it is a problem in the point of cpu and memory are requested well with that combination.

I can see a todo in the code base noting this issue. We will get this addressed and make a release next Monday. Thanks for pointing this out!

Thanks for your quick response!