I am using Openfaas to run some tasks on the GPU. However, since these tasks use only the GPU, and the CPU remains mostly idle, openfaas does not scale these pods, and hence GKE does not scale our nodes. How can I configure openfaas to use GPU metrics for scaling?

My understanding is that CPU usage is currently being sent to prometheus, which is then being used by alert manager to scale the pods up or down. I need to somehow make this such that whenever a pod uses more than 10% GPU for 1 second then scale it up.