Use vLLM on GKE to run inference with gpt-oss-120b

This tutorial shows you how to deploy and serve a gpt-oss-120b, language model by using the vLLM framework. You deploy this model on a Google Kubernetes Engine (GKE) autopilot cluster and consume a single A4 virtual machine (VM) that has 8 B200 GPUs.

This tutorial is intended for machine learning (ML) engineers, platform administrators and operators, and for data and AI specialists who are interested in using Kubernetes container orchestration capabilities to handle inference workloads.

Access gpt-oss by using Hugging Face

To use Hugging Face to access gpt-oss, do the following:

  1. Sign in to Hugging Face and explore the gpt-oss model.
  2. Create a Hugging Face read access token.
  3. Copy and save the read access token value. You use it later in this tutorial.

Prepare your environment

To prepare your environment, set the default environment variables:

gcloud config set project PROJECT_ID
gcloud config set billing/quota_project PROJECT_ID
export PROJECT_ID=$(gcloud config get project)
export RESERVATION_URL=RESERVATION_URL
export REGION=REGION
export CLUSTER_NAME=CLUSTER_NAME
export HUGGING_FACE_TOKEN=HUGGING_FACE_TOKEN
export NETWORK=NETWORK_NAME
export SUBNETWORK=SUBNETWORK_NAME

Replace the following:

  • PROJECT_ID: the ID of the Google Cloud project where you want to create the GKE cluster.

  • RESERVATION_URL: the URL of the reservation that you want to use to create your GKE cluster. Based on the project in which the reservation exists, specify one of the following values:

    • The reservation exists in your project: RESERVATION_NAME

    • The reservation exists in a different project, and your project can use the reservation: projects/RESERVATION_PROJECT_ID/reservations/RESERVATION_NAME

  • REGION: the region where you want to create your GKE cluster. You can only create the cluster in the region where your reservation exists.

  • CLUSTER_NAME: the name of the GKE cluster to create.

  • HUGGING_FACE_TOKEN: the Hugging Face access token that you created in the previous section.

  • NETWORK_NAME: the network that the GKE cluster uses. Specify one of the following values:

    • If you created a custom network, then specify the name of your network.

    • Otherwise, specify default.

  • SUBNETWORK_NAME: the subnetwork that the GKE cluster uses. Specify one of the following values:

    • If you created a custom subnetwork, then specify the name of your subnetwork. You can only specify a subnetwork that exists in the same region as the reservation.

    • Otherwise, specify default.

Create a GKE cluster in Autopilot mode

To create a GKE cluster in Autopilot mode, run the following command:

gcloud container clusters create-auto $CLUSTER_NAME \
    --project=$PROJECT_ID \
    --region=$REGION \
    --release-channel=rapid \
    --network=$NETWORK \
    --subnetwork=$SUBNETWORK

Creating the GKE cluster might take some time to complete. To verify that Google Cloud has finished creating your cluster, go to Kubernetes clusters on the Google Cloud console.

Create a Kubernetes secret for Hugging Face credentials

To create a Kubernetes secret for Hugging Face credentials, do the following:

  1. Configure kubectl to communicate with your GKE cluster:

    gcloud container clusters get-credentials $CLUSTER_NAME \
    --location=$REGION

  2. Create a Kubernetes secret to store your Hugging Face token:

    kubectl create secret generic hf-secret \
    --from-literal=hf_token=${HUGGING_FACE_TOKEN} \
    --dry-run=client -o yaml | kubectl apply -f -

Deploy a vLLM container to your GKE cluster

  1. Create a vllm-gpt-oss-120b.yaml file with your chosen vLLM deployment:

    apiVersion: apps/v1
kind: Deployment
metadata:
  name: vllm-gpt-oss-deployment
spec:
  replicas: 1
  selector:
    matchLabels:
      app: gpt-oss
  template:
    metadata:
      labels:
        app: gpt-oss
        ai.gke.io/model: gpt-oss-120b
        ai.gke.io/inference-server: vllm
        examples.ai.gke.io/source: user-guide
    spec:
      containers:
      - name: vllm-inference
        image: us-docker.pkg.dev/vertex-ai/vertex-vision-model-garden-dockers/pytorch-vllm-serve:20250822_0916_RC01
        resources:
          requests:
            cpu: "10"
            memory: "128Gi"
            ephemeral-storage: "240Gi"
            nvidia.com/gpu: "8"
          limits:
            cpu: "10"
            memory: "128Gi"
            ephemeral-storage: "240Gi"
            nvidia.com/gpu: "8"
        command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
        args:
        - --model=$(MODEL_ID)
        - --tensor-parallel-size=2
        - --host=0.0.0.0
        - --port=8000
        - --max-model-len=8192
        - --max-num-seqs=4
        env:
        - name: MODEL_ID
          value: openai/gpt-oss-120b
        - name: HUGGING_FACE_HUB_TOKEN
          valueFrom:
            secretKeyRef:
              name: hf-secret
              key: hf_token
        volumeMounts:
        - mountPath: /dev/shm
          name: dshm
        livenessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 1200
          periodSeconds: 10
        readinessProbe:
          httpGet:
            path: /health
            port: 8000
          initialDelaySeconds: 1200
          periodSeconds: 5
      volumes:
      - name: dshm
        emptyDir:
            medium: Memory
      nodeSelector:
        cloud.google.com/gke-accelerator: nvidia-b200
        cloud.google.com/reservation-name: $RESERVATION_URL
        cloud.google.com/reservation-affinity: "specific"
        cloud.google.com/gke-gpu-driver-version: latest
---
apiVersion: v1
kind: Service
metadata:
  name: oss-service
spec:
  selector:
    app: gpt-oss
  type: ClusterIP
  ports:
    - protocol: TCP
      port: 8000
      targetPort: 8000
---
apiVersion: monitoring.googleapis.com/v1
kind: PodMonitoring
metadata:
  name: vllm-gpt-oss-monitoring
spec:
  selector:
    matchLabels:
      app: gpt-oss
  endpoints:
  - port: 8000
    path: /metrics
    interval: 30s

  2. Apply the vllm-gpt-oss-120b.yaml file to your GKE cluster:

    envsubst < vllm-gpt-oss-120b.yaml | kubectl apply -f -

  3. During the deployment process, the container must download the gpt-oss-120b model from Hugging Face. For this reason, deployment of the container might take up to 20 minutes to complete.

  4. To see the completion status, run the following command:

    kubectl wait \
--for=condition=Available \
--timeout=1200s deployment/vllm-gpt-oss-deployment

    The --timeout=1200s flag allows the command to monitor the deployment for up to 20 minutes.

Interact with the gpt-oss model by using curl

To verify the gpt-oss model that you deployed, do the following:

  1. Set up port forwarding to the gpt-oss model:

    kubectl port-forward service/oss-service 8000:8000

  2. Open a new terminal window. You can then chat with your model by usingcurl:

    curl http://127.0.0.1:8000/v1/chat/completions \
-X POST \
-H "Content-Type: application/json" \
-d '{
  "model": "openai/gpt-oss-120b",
  "messages": [
    {
      "role": "user",
      "content": "Describe a sailboat in one short sentence?"
    }
  ]
}'

  3. The output that you see is similar to the following:

    {
  "id": "chatcmpl-2235c39759c040daae23ce2addc40c0a",
  "object": "chat.completion",
  "created": 1756831629,
  "model": "openai/gpt-oss-120b",
  "choices": [
    {
      "index": 0,
      "message": {
        "role": "assistant",
        "content": "A sleek vessel gliding on water, its cloth sails billowing like captured wind.",
        "refusal": null,
        "annotations": null,
        "audio": null,
        "function_call": null,
        "tool_calls": [],
        "reasoning_content": "User asks: \"Describe a sailboat in one short sentence?\" We need to produce a short sentence description. Should comply with policy. It's fine. Provide a short sentence."
      },
      "logprobs": null,
      "finish_reason": "stop",
      "stop_reason": null
    }
  ],
  "service_tier": null,
  "system_fingerprint": null,
  "usage": {
    "prompt_tokens": 80,
    "total_tokens": 142,
    "completion_tokens": 62,
    "prompt_tokens_details": null
  },
  "prompt_logprobs": null,
  "kv_transfer_params": null
}

    Observe the performance of the model

To observe your model's performance, you can use the vLLM dashboard integration in Cloud Monitoring. This dashboard helps you view critical performance metrics for your model like token throughput, network latency, and error rates. For information, see vLLM in the Monitoring documentation.