apiVersion: apps.nvidia.com/v1alpha1
kind: NIMCache
metadata:
  name: qwen3-32b-instruct
  namespace: nim-service
spec:
  source:
    ngc:
      modelPuller: nvcr.io/nim/qwen/qwen3-32b-dgx-spark:1.1.0-variant
      pullSecret: ngc-secret
      authSecret: ngc-api-secret
      model:
        engine: "vllm"
        tensorParallelism: "1"
        profiles:
          - c4f105d92c72ab56200884dfacde9d2128b139755c06b9c883eeb3e287b7408a
  storage:
    pvc:
      create: true
      size: "100Gi"
      volumeAccessMode: ReadWriteOnce
---
apiVersion: apps.nvidia.com/v1alpha1
kind: NIMService
metadata:
  name: qwen3-32b-instruct
  namespace: nim-service
spec:
  env:
    - name: GPU_MEMORY_UTILIZATION
      value: "0.90"
    - name: NIM_MAX_NUM_SEQS
      value: "32"
    - name: NIM_MAX_NUM_BATCHED_TOKENS
      value: "16384"
    - name: ENABLE_AUTO_TOOL_CHOICE
      value: "true"
    - name: ENABLE_PREFIX_CACHING
      value: "true"
    - name: TRUNCATION_SIDE
      value: "left"
    - name: VLLM_LOGGING_LEVEL
      value: "INFO"
    - name: VLLM_ALLOW_LONG_MAX_MODEL_LEN
      value: "true"
    - name: NIM_MAX_MODEL_LEN
      value: "32768"
    - name: NIM_ENABLE_KV_CACHE_REUSE
      value: "1"
  image:
    repository: nvcr.io/nim/qwen/qwen3-32b-dgx-spark
    tag: "1.1.0-variant"
    pullPolicy: IfNotPresent
    pullSecrets:
      - ngc-secret
  authSecret: ngc-api-secret
  storage:
    nimCache:
      name: qwen3-32b-instruct
  replicas: 1
  resources:
    limits:
      nvidia.com/gpu: 1
  expose:
    service:
      type: ClusterIP
      port: 8000
---
apiVersion: gateway.networking.k8s.io/v1
kind: HTTPRoute
metadata:
  name: llm-route
  namespace: nim-service
spec:
  parentRefs:
    - name: envoy-gateway
      namespace: default
  hostnames:
    - "mcp.corredorconect.com"
  rules:
    - matches:
        - path:
            type: PathPrefix
            value: /
      backendRefs:
        - name: qwen3-32b-instruct
          port: 8000