Allocate specific QoS policy devices

Users can configure the QoS policy for tasks using the metax-tech.com/sgpu-qos-policy annotation to specify the scheduling policy used by the shared GPU (sGPU). The available sGPU scheduling policies are described in the table below:

Scheduling Policy	Description
best-effort	The sGPU has no restriction on compute usage.
fixed-share	The sGPU is assigned a fixed compute quota and cannot exceed this limit.
burst-share	The sGPU is assigned a fixed compute quota, but may utilize additional GPU compute resources when they are idle.

apiVersion: v1
kind: Pod
metadata:
  name: gpu-pod
  annotations:
    metax-tech.com/sgpu-qos-policy: "best-effort" # allocate specific qos sgpu
spec:
  containers:
    - name: ubuntu-container
      image: ubuntu:22.04
      imagePullPolicy: IfNotPresent
      command: ["sleep","infinity"]
      resources:
        limits:
          metax-tech.com/sgpu: 1 # requesting 1 GPU
          metax-tech.com/vcore: 60 # each GPU use 60% of total compute cores
          metax-tech.com/vmemory: 4 # each GPU require 4 GiB device memory