# GPU platform configuration for OSMO.
# Edit this file when adding/removing GPU node pools.
# Must match Terraform node pool configuration (vm_size, taints).
services:
  configs:
    podTemplates:
      gpu_tpl:
        spec:
          nodeSelector:
            node.kubernetes.io/instance-type: "Standard_NC24ads_A100_v4"
          tolerations:
            - key: "nvidia.com/gpu"
              operator: "Exists"
              effect: "NoSchedule"
            - key: "kubernetes.azure.com/scalesetpriority"
              value: "spot"
              operator: "Equal"
              effect: "NoSchedule"
          containers:
            - name: "{{USER_CONTAINER_NAME}}"
              resources:
                requests: { nvidia.com/gpu: "{{USER_GPU}}" }
                limits: { nvidia.com/gpu: "{{USER_GPU}}" }
      # Multi-GPU node pool (2x A100 per node). Single-node multi-GPU jobs
      # (USER_GPU > 1) must target this SKU; the 1x SKU above cannot satisfy a
      # 2-GPU request because each NC24 node exposes only one GPU.
      gpu_tpl_2x:
        spec:
          nodeSelector:
            node.kubernetes.io/instance-type: "Standard_NC48ads_A100_v4"
          tolerations:
            - key: "nvidia.com/gpu"
              operator: "Exists"
              effect: "NoSchedule"
            - key: "kubernetes.azure.com/scalesetpriority"
              value: "spot"
              operator: "Equal"
              effect: "NoSchedule"
          containers:
            - name: "{{USER_CONTAINER_NAME}}"
              resources:
                requests: { nvidia.com/gpu: "{{USER_GPU}}" }
                limits: { nvidia.com/gpu: "{{USER_GPU}}" }

    resourceValidations:
      gpu_gpu_required:
        - { operator: GE, left_operand: "{{USER_GPU}}", right_operand: "1", assert_message: "GPU platform: GPU must be >= 1" }

    pools:
      default:
        platforms:
          gpu_platform:
            override_pod_template: [gpu_tpl]
            resource_validations: [gpu_gpu_required]
            default_variables:
              USER_GPU: 1
              USER_CPU: 8
              USER_MEMORY: "64Gi"
              USER_STORAGE: "100Gi"
              USER_SHM_SIZE: "16Gi"
          # Single-node multi-GPU platform: binds gpu_tpl_2x (2x A100 SKU).
          # Select it for jobs requesting more than one GPU.
          gpu_platform_2x:
            override_pod_template: [gpu_tpl_2x]
            resource_validations: [gpu_gpu_required]
            default_variables:
              USER_GPU: 2
              USER_CPU: 8
              USER_MEMORY: "64Gi"
              USER_STORAGE: "100Gi"
              USER_SHM_SIZE: "16Gi"