Hi, i'm currently setting up the GCP autoscaler. Everything works except the installation of the cuda drivers. I'm using this machine image projects/ml-images/global/images/c0-deeplearning-common-cu113-v20240319-debian-11-py310 that provides a install script /opt/deeplearning/install-driver.sh . When executed this error occurs:

ERROR: (gcloud.projects.describe) User [XXXX@developer.gserviceaccount.com] does not have permission to access projects instance [XXXX] (or it may not exist): Request had insufficient authentication scopes.
- '@type': type.googleapis.com/google.rpc.ErrorInfo
  domain: googleapis.com
  metadata:
    method: google.cloudresourcemanager.v1.Projects.GetProject
    service: cloudresourcemanager.googleapis.com
  reason: ACCESS_TOKEN_SCOPE_INSUFFICIENT
If you are in a compute engine VM, it is likely that the specified scopes during VM creation are not enough to run this command.
See 

 for more information about access scopes.
See 

 for how to update access scopes of the VM.

As I understand it a possible solution would be to set the scope None when spinning up the machine. I don't think there's a setting for that. Also I think this thread is related: None
Any ideas?