透過 GKE 推論閘道提供 LLM


本教學課程說明如何使用 GKE Inference Gateway,在 Google Kubernetes Engine (GKE) 上部署大型語言模型 (LLM)。本教學課程包含叢集設定、模型部署、GKE Inference Gateway 設定,以及處理 LLM 要求等步驟。

本教學課程適用於機器學習 (ML) 工程師、平台管理員和營運人員,以及資料和 AI 專家。他們想使用 GKE Inference Gateway,在 GKE 上部署及管理 LLM 應用程式。

閱讀本頁內容前,請先熟悉下列概念:

背景

本節說明本教學課程中使用的主要技術。如要進一步瞭解模型服務概念和術語,以及 GKE 生成式 AI 功能如何提升及支援模型服務效能,請參閱「關於 GKE 上的模型推論」。

vLLM

vLLM 是經過高度最佳化的開放原始碼 LLM 服務架構,可提高 GPU 的服務輸送量,並提供下列功能:

  • 使用 PagedAttention 實作最佳化轉換器
  • 持續批次處理,提升整體服務輸送量
  • 張量平行處理和分散式服務,可跨多個 GPU 運作

詳情請參閱 vLLM 說明文件

GKE 推論閘道

GKE 推論閘道可強化 GKE 的功能,以便提供 LLM 服務。這項服務可透過下列功能,將推論工作負載最佳化:

  • 根據負載指標,以推論最佳化負載平衡。
  • 支援 LoRA 適應器的密集多工作負載服務。
  • 可感知模型的路由,簡化作業。

詳情請參閱「關於 GKE Inference Gateway」。

目標

  1. 取得模型存取權
  2. 準備環境
  3. 建立及設定 Google Cloud 資源
  4. 安裝 InferenceModelInferencePool CRD
  5. 部署模型伺服器
  6. 設定 Inference Gateway 的可觀測性

事前準備

  • Sign in to your Google Cloud account. If you're new to Google Cloud, create an account to evaluate how our products perform in real-world scenarios. New customers also get $300 in free credits to run, test, and deploy workloads.
  • In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

    Go to project selector

  • Make sure that billing is enabled for your Google Cloud project.

  • Enable the required API.

    Enable the API

  • In the Google Cloud console, on the project selector page, select or create a Google Cloud project.

    Go to project selector

  • Make sure that billing is enabled for your Google Cloud project.

  • Enable the required API.

    Enable the API

  • Make sure that you have the following role or roles on the project: roles/container.admin, roles/iam.serviceAccountAdmin

    Check for the roles

    1. In the Google Cloud console, go to the IAM page.

      Go to IAM
    2. Select the project.
    3. In the Principal column, find all rows that identify you or a group that you're included in. To learn which groups you're included in, contact your administrator.

    4. For all rows that specify or include you, check the Role column to see whether the list of roles includes the required roles.

    Grant the roles

    1. In the Google Cloud console, go to the IAM page.

      前往「IAM」頁面
    2. 選取專案。
    3. 按一下「授予存取權」
    4. 在「New principals」(新增主體) 欄位中,輸入您的使用者 ID。 這通常是 Google 帳戶的電子郵件地址。

    5. 在「Select a role」(選取角色) 清單中,選取角色。
    6. 如要授予其他角色,請按一下 「新增其他角色」,然後新增每個其他角色。
    7. 按一下 [Save]

取得模型存取權

如要將 Llama3.1 模型部署至 GKE,請簽署授權同意聲明協議,並產生 Hugging Face 存取權杖。

您必須簽署同意聲明,才能使用 Llama3.1 模型。請按照以下步驟操作:

  1. 前往同意聲明頁面,確認您同意使用 Hugging Face 帳戶。
  2. 接受模型條款。

產生存取權杖

如要透過 Hugging Face 存取模型,您需要 Hugging Face 權杖

如要產生新權杖 (如果沒有),請按照下列步驟操作:

  1. 依序點選「Your Profile」(你的個人資料) >「Settings」(設定) >「Access Tokens」(存取權杖)
  2. 選取「New Token」
  3. 指定所選名稱和至少 Read 的角色。
  4. 選取「產生權杖」
  5. 將產生的權杖複製到剪貼簿。

準備環境

在本教學課程中,您將使用 Cloud Shell 管理託管於Google Cloud的資源。Cloud Shell 已預先安裝本教學課程所需的軟體,包括 kubectl gcloud CLI

如要使用 Cloud Shell 設定環境,請執行下列步驟:

  1. 在 Google Cloud 控制台中,按一下Cloud Shell 啟用圖示Google Cloud 控制台中的「啟用 Cloud Shell」,啟動 Cloud Shell 工作階段。系統會在 Google Cloud 控制台的底部窗格啟動工作階段。

  2. 設定預設環境變數:

    gcloud config set project PROJECT_ID
    gcloud config set billing/quota_project PROJECT_ID
    export PROJECT_ID=$(gcloud config get project)
    export REGION=REGION
    export CLUSTER_NAME=CLUSTER_NAME
    export HF_TOKEN=HF_TOKEN
    

    替換下列值:

    • PROJECT_ID:您的 Google Cloud 專案 ID
    • REGION:支援您要使用的加速器類型,例如 H100 GPU 的 us-central1
    • CLUSTER_NAME:叢集名稱。
    • HF_TOKEN:您先前產生的 Hugging Face 權杖。

建立及設定 Google Cloud 資源

如要建立必要資源,請按照這些操作說明進行。

建立 GKE 叢集和節點集區

在 GKE Autopilot 或 Standard 叢集的 GPU 上提供 LLM。建議您使用 Autopilot 叢集,享受全代管 Kubernetes 體驗。如要為工作負載選擇最合適的 GKE 作業模式,請參閱「選擇 GKE 作業模式」。

Autopilot

在 Cloud Shell 中執行下列指令:

gcloud container clusters create-auto CLUSTER_NAME \
    --project=PROJECT_ID \
    --region=REGION \
    --release-channel=rapid

替換下列值:

  • PROJECT_ID:您的 Google Cloud 專案 ID
  • REGION:支援您要使用的加速器類型,例如 H100 GPU 的 us-central1
  • CLUSTER_NAME:叢集名稱。

GKE 會根據部署的工作負載要求,建立含 CPU 和 GPU 節點的 Autopilot 叢集。

標準

  1. 在 Cloud Shell 中執行下列指令,建立 Standard 叢集:

    gcloud container clusters create CLUSTER_NAME \
        --project=PROJECT_ID \
        --region=REGION \
        --workload-pool=PROJECT_ID.svc.id.goog \
        --release-channel=rapid \
        --num-nodes=1 \
        --enable-managed-prometheus \
        --monitoring=SYSTEM,DCGM
    

    替換下列值:

    • PROJECT_ID:您的 Google Cloud 專案 ID
    • REGION:支援您要使用的加速器類型,例如 H100 GPU 的 us-central1
    • CLUSTER_NAME:叢集名稱。

    建立叢集可能需要幾分鐘的時間。

  2. 如要建立節點集區,並為執行 Llama-3.1-8B-Instruct 模型設定適當的磁碟大小,請執行下列指令:

    gcloud container node-pools create gpupool \
        --accelerator type=nvidia-h100-80gb,count=2,gpu-driver-version=latest \
        --project=PROJECT_ID \
        --location=REGION \
        --node-locations=REGION-a \
        --cluster=CLUSTER_NAME \
        --machine-type=a3-highgpu-2g \
        --num-nodes=1 \
        --disk-type="pd-standard"
    

    GKE 會建立包含 H100 GPU 的單一節點集區。

  1. 如要設定授權來擷取指標,請建立 inference-gateway-sa-metrics-reader-secret 密鑰:

    kubectl apply -f - <<EOF
    ---
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRole
    metadata:
      name: inference-gateway-metrics-reader
    rules:
    - nonResourceURLs:
      - /metrics
      verbs:
      - get
    ---
    apiVersion: v1
    kind: ServiceAccount
    metadata:
      name: inference-gateway-sa-metrics-reader
      namespace: default
    ---
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRoleBinding
    metadata:
      name: inference-gateway-sa-metrics-reader-role-binding
      namespace: default
    subjects:
    - kind: ServiceAccount
      name: inference-gateway-sa-metrics-reader
      namespace: default
    roleRef:
      kind: ClusterRole
      name: inference-gateway-metrics-reader
      apiGroup: rbac.authorization.k8s.io
    ---
    apiVersion: v1
    kind: Secret
    metadata:
      name: inference-gateway-sa-metrics-reader-secret
      namespace: default
      annotations:
        kubernetes.io/service-account.name: inference-gateway-sa-metrics-reader
    type: kubernetes.io/service-account-token
    ---
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRole
    metadata:
      name: inference-gateway-sa-metrics-reader-secret-read
    rules:
    - resources:
      - secrets
      apiGroups: [""]
      verbs: ["get", "list", "watch"]
      resourceNames: ["inference-gateway-sa-metrics-reader-secret"]
    ---
    apiVersion: rbac.authorization.k8s.io/v1
    kind: ClusterRoleBinding
    metadata:
      name: gmp-system:collector:inference-gateway-sa-metrics-reader-secret-read
      namespace: default
    roleRef:
      name: inference-gateway-sa-metrics-reader-secret-read
      kind: ClusterRole
      apiGroup: rbac.authorization.k8s.io
    subjects:
    - name: collector
      namespace: gmp-system
      kind: ServiceAccount
    EOF
    

為 Hugging Face 憑證建立 Kubernetes 密鑰

在 Cloud Shell 中執行下列操作:

  1. 如要與叢集通訊,請設定 kubectl

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

    替換下列值:

    • REGION:支援您要使用的加速器類型,例如 H100 GPU 的 us-central1
    • CLUSTER_NAME:叢集名稱。
  2. 建立包含 Hugging Face 權杖的 Kubernetes Secret:

      kubectl create secret generic HF_SECRET \
          --from-literal=token=HF_TOKEN \
          --dry-run=client -o yaml | kubectl apply -f -
    

    更改下列內容:

    • HF_TOKEN:您先前產生的 Hugging Face 權杖。
    • HF_SECRET:Kubernetes Secret 的名稱。例如:hf-secret

安裝 InferenceModelInferencePool CRD

在本節中,您將安裝 GKE Inference Gateway 的必要自訂資源定義 (CRD)。

CRD 會擴充 Kubernetes API,這可讓您定義新的資源類型。如要使用 GKE Inference Gateway,請在 GKE 叢集中安裝 InferencePoolInferenceModel CRD,方法是執行下列指令:

kubectl apply -f https://github.com/kubernetes-sigs/gateway-api-inference-extension/releases/download/v0.3.0/manifests.yaml

部署模型伺服器

這個範例會使用 vLLM 模型伺服器部署 Llama3.1 模型。部署作業標示為 app:vllm-llama3-8b-instruct。這項部署作業也會使用來自 Hugging Face 的兩個 LoRA 適應器,分別命名為 food-reviewcad-fabricator。您可以更新這項部署作業,加入自己的模型伺服器和模型容器、服務通訊埠和部署作業名稱。您可以選擇在部署作業中設定 LoRA 配接器,或部署基礎模型。

  1. 如要在 nvidia-h100-80gb 加速器類型上部署,請將下列資訊清單儲存為 vllm-llama3-8b-instruct.yaml。這個資訊清單會定義含有模型和模型伺服器的 Kubernetes 部署:

    apiVersion: apps/v1
    kind: Deployment
    metadata:
      name: vllm-llama3-8b-instruct
    spec:
      replicas: 3
      selector:
        matchLabels:
          app: vllm-llama3-8b-instruct
      template:
        metadata:
          labels:
            app: vllm-llama3-8b-instruct
        spec:
          containers:
            - name: vllm
              image: "vllm/vllm-openai:latest"
              imagePullPolicy: Always
              command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
              args:
              - "--model"
              - "meta-llama/Llama-3.1-8B-Instruct"
              - "--tensor-parallel-size"
              - "1"
              - "--port"
              - "8000"
              - "--enable-lora"
              - "--max-loras"
              - "2"
              - "--max-cpu-loras"
              - "12"
              env:
                # Enabling LoRA support temporarily disables automatic v1, we want to force it on
                # until 0.8.3 vLLM is released.
                - name: VLLM_USE_V1
                  value: "1"
                - name: PORT
                  value: "8000"
                - name: HUGGING_FACE_HUB_TOKEN
                  valueFrom:
                    secretKeyRef:
                      name: hf-token
                      key: token
                - name: VLLM_ALLOW_RUNTIME_LORA_UPDATING
                  value: "true"
              ports:
                - containerPort: 8000
                  name: http
                  protocol: TCP
              lifecycle:
                preStop:
                  # vLLM stops accepting connections when it receives SIGTERM, so we need to sleep
                  # to give upstream gateways a chance to take us out of rotation. The time we wait
                  # is dependent on the time it takes for all upstreams to completely remove us from
                  # rotation. Older or simpler load balancers might take upwards of 30s, but we expect
                  # our deployment to run behind a modern gateway like Envoy which is designed to
                  # probe for readiness aggressively.
                  sleep:
                    # Upstream gateway probers for health should be set on a low period, such as 5s,
                    # and the shorter we can tighten that bound the faster that we release
                    # accelerators during controlled shutdowns. However, we should expect variance,
                    # as load balancers may have internal delays, and we don't want to drop requests
                    # normally, so we're often aiming to set this value to a p99 propagation latency
                    # of readiness -> load balancer taking backend out of rotation, not the average.
                    #
                    # This value is generally stable and must often be experimentally determined on
                    # for a given load balancer and health check period. We set the value here to
                    # the highest value we observe on a supported load balancer, and we recommend
                    # tuning this value down and verifying no requests are dropped.
                    #
                    # If this value is updated, be sure to update terminationGracePeriodSeconds.
                    #
                    seconds: 30
                  #
                  # IMPORTANT: preStop.sleep is beta as of Kubernetes 1.30 - for older versions
                  # replace with this exec action.
                  #exec:
                  #  command:
                  #  - /usr/bin/sleep
                  #  - 30
              livenessProbe:
                httpGet:
                  path: /health
                  port: http
                  scheme: HTTP
                # vLLM's health check is simple, so we can more aggressively probe it.  Liveness
                # check endpoints should always be suitable for aggressive probing.
                periodSeconds: 1
                successThreshold: 1
                # vLLM has a very simple health implementation, which means that any failure is
                # likely significant. However, any liveness triggered restart requires the very
                # large core model to be reloaded, and so we should bias towards ensuring the
                # server is definitely unhealthy vs immediately restarting. Use 5 attempts as
                # evidence of a serious problem.
                failureThreshold: 5
                timeoutSeconds: 1
              readinessProbe:
                httpGet:
                  path: /health
                  port: http
                  scheme: HTTP
                # vLLM's health check is simple, so we can more aggressively probe it.  Readiness
                # check endpoints should always be suitable for aggressive probing, but may be
                # slightly more expensive than readiness probes.
                periodSeconds: 1
                successThreshold: 1
                # vLLM has a very simple health implementation, which means that any failure is
                # likely significant,
                failureThreshold: 1
                timeoutSeconds: 1
              # We set a startup probe so that we don't begin directing traffic or checking
              # liveness to this instance until the model is loaded.
              startupProbe:
                # Failure threshold is when we believe startup will not happen at all, and is set
                # to the maximum possible time we believe loading a model will take. In our
                # default configuration we are downloading a model from HuggingFace, which may
                # take a long time, then the model must load into the accelerator. We choose
                # 10 minutes as a reasonable maximum startup time before giving up and attempting
                # to restart the pod.
                #
                # IMPORTANT: If the core model takes more than 10 minutes to load, pods will crash
                # loop forever. Be sure to set this appropriately.
                failureThreshold: 3600
                # Set delay to start low so that if the base model changes to something smaller
                # or an optimization is deployed, we don't wait unnecessarily.
                initialDelaySeconds: 2
                # As a startup probe, this stops running and so we can more aggressively probe
                # even a moderately complex startup - this is a very important workload.
                periodSeconds: 1
                httpGet:
                  # vLLM does not start the OpenAI server (and hence make /health available)
                  # until models are loaded. This may not be true for all model servers.
                  path: /health
                  port: http
                  scheme: HTTP
    
              resources:
                limits:
                  nvidia.com/gpu: 1
                requests:
                  nvidia.com/gpu: 1
              volumeMounts:
                - mountPath: /data
                  name: data
                - mountPath: /dev/shm
                  name: shm
                - name: adapters
                  mountPath: "/adapters"
          initContainers:
            - name: lora-adapter-syncer
              tty: true
              stdin: true
              image: us-central1-docker.pkg.dev/k8s-staging-images/gateway-api-inference-extension/lora-syncer:main
              restartPolicy: Always
              imagePullPolicy: Always
              env:
                - name: DYNAMIC_LORA_ROLLOUT_CONFIG
                  value: "/config/configmap.yaml"
              volumeMounts: # DO NOT USE subPath, dynamic configmap updates don't work on subPaths
              - name: config-volume
                mountPath:  /config
          restartPolicy: Always
    
          # vLLM allows VLLM_PORT to be specified as an environment variable, but a user might
          # create a 'vllm' service in their namespace. That auto-injects VLLM_PORT in docker
          # compatible form as `tcp://<IP>:<PORT>` instead of the numeric value vLLM accepts
          # causing CrashLoopBackoff. Set service environment injection off by default.
          enableServiceLinks: false
    
          # Generally, the termination grace period needs to last longer than the slowest request
          # we expect to serve plus any extra time spent waiting for load balancers to take the
          # model server out of rotation.
          #
          # An easy starting point is the p99 or max request latency measured for your workload,
          # although LLM request latencies vary significantly if clients send longer inputs or
          # trigger longer outputs. Since steady state p99 will be higher than the latency
          # to drain a server, you may wish to slightly this value either experimentally or
          # via the calculation below.
          #
          # For most models you can derive an upper bound for the maximum drain latency as
          # follows:
          #
          #   1. Identify the maximum context length the model was trained on, or the maximum
          #      allowed length of output tokens configured on vLLM (llama2-7b was trained to
          #      4k context length, while llama3-8b was trained to 128k).
          #   2. Output tokens are the more compute intensive to calculate and the accelerator
          #      will have a maximum concurrency (batch size) - the time per output token at
          #      maximum batch with no prompt tokens being processed is the slowest an output
          #      token can be generated (for this model it would be about 100ms TPOT at a max
          #      batch size around 50)
          #   3. Calculate the worst case request duration if a request starts immediately
          #      before the server stops accepting new connections - generally when it receives
          #      SIGTERM (for this model that is about 4096 / 10 ~ 40s)
          #   4. If there are any requests generating prompt tokens that will delay when those
          #      output tokens start, and prompt token generation is roughly 6x faster than
          #      compute-bound output token generation, so add 20% to the time from above (40s +
          #      16s ~ 55s)
          #
          # Thus we think it will take us at worst about 55s to complete the longest possible
          # request the model is likely to receive at maximum concurrency (highest latency)
          # once requests stop being sent.
          #
          # NOTE: This number will be lower than steady state p99 latency since we stop       receiving
          #       new requests which require continuous prompt token computation.
              # NOTE: The max timeout for backend connections from gateway to model servers should
          #       be configured based on steady state p99 latency, not drain p99 latency
          #
          #   5. Add the time the pod takes in its preStop hook to allow the load balancers have
          #      stopped sending us new requests (55s + 30s ~ 85s)
          #
          # Because the termination grace period controls when the Kubelet forcibly terminates a
          # stuck or hung process (a possibility due to a GPU crash), there is operational safety
          # in keeping the value roughly proportional to the time to finish serving. There is also
          # value in adding a bit of extra time to deal with unexpectedly long workloads.
          #
          #   6. Add a 50% safety buffer to this time since the operational impact should be low
          #      (85s * 1.5 ~ 130s)
          #
          # One additional source of drain latency is that some workloads may run close to
          # saturation and have queued requests on each server. Since traffic in excess of the
          # max sustainable QPS will result in timeouts as the queues grow, we assume that failure
          # to drain in time due to excess queues at the time of shutdown is an expected failure
          # mode of server overload. If your workload occasionally experiences high queue depths
          # due to periodic traffic, consider increasing the safety margin above to account for
          # time to drain queued requests.
          terminationGracePeriodSeconds: 130
          nodeSelector:
            cloud.google.com/gke-accelerator: "nvidia-h100-80gb"
          volumes:
            - name: data
              emptyDir: {}
            - name: shm
              emptyDir:
                medium: Memory
            - name: adapters
              emptyDir: {}
            - name: config-volume
              configMap:
                name: vllm-llama3-8b-adapters
    ---
    apiVersion: v1
    kind: ConfigMap
    metadata:
      name: vllm-llama3-8b-adapters
    data:
      configmap.yaml: |
          vLLMLoRAConfig:
            name: vllm-llama3.1-8b-instruct
            port: 8000
            defaultBaseModel: meta-llama/Llama-3.1-8B-Instruct
            ensureExist:
              models:
              - id: food-review
                source: Kawon/llama3.1-food-finetune_v14_r8
              - id: cad-fabricator
                source: redcathode/fabricator
    ---
    kind: HealthCheckPolicy
    apiVersion: networking.gke.io/v1
    metadata:
      name: health-check-policy
      namespace: default
    spec:
      targetRef:
        group: "inference.networking.x-k8s.io"
        kind: InferencePool
        name: vllm-llama3-8b-instruct
      default:
        config:
          type: HTTP
          httpHealthCheck:
              requestPath: /health
              port: 8000
    
  2. 將資訊清單套用至叢集:

    kubectl apply -f vllm-llama3-8b-instruct.yaml
    

建立 InferencePool 資源

InferencePool Kubernetes 自訂資源會定義一組 Pod,這些 Pod 具有共同的基礎 LLM 和運算設定。

InferencePool 自訂資源包含下列重要欄位:

  • selector:指定哪些 Pod 屬於這個集區。這個選取器中的標籤必須與套用至模型伺服器 Pod 的標籤完全一致。
  • targetPort:定義 Pod 中模型伺服器使用的通訊埠。

InferencePool 資源可讓 GKE 推論閘道將流量轉送至模型伺服器 Pod。

如要使用 Helm 建立 InferencePool,請執行下列步驟:

helm install vllm-llama3-8b-instruct \
  --set inferencePool.modelServers.matchLabels.app=vllm-llama3-8b-instruct \
  --set provider.name=gke \
  --set healthCheckPolicy.create=false \
  --version v0.3.0 \
  oci://registry.k8s.io/gateway-api-inference-extension/charts/inferencepool

變更下列欄位,以符合您的部署作業:

  • inferencePool.modelServers.matchLabels.app:用於選取模型伺服器 Pod 的標籤鍵。

這個指令會建立 InferencePool 物件,以邏輯方式代表模型伺服器部署作業,並參照 Selector 選取的 Pod 內模型端點服務。

建立具有服務重要性的 InferenceModel 資源

InferenceModel Kubernetes 自訂資源會定義特定模型,包括 LoRA 微調模型,以及其服務重要性。

InferenceModel 自訂資源包含下列重要欄位:

  • modelName:指定基礎模型或 LoRA 適應器的名稱。
  • Criticality:指定模型的服務重要性。
  • poolRef:參照模型服務的 InferencePool

InferenceModel 可讓 GKE Inference Gateway 根據模型名稱和重要性,將流量導向至模型伺服器 Pod。

如要建立 InferenceModel,請執行下列步驟:

  1. 將下列範例資訊清單儲存為 inferencemodel.yaml

    apiVersion: inference.networking.x-k8s.io/v1alpha2
    kind: InferenceModel
    metadata:
      name: inferencemodel-sample
    spec:
      modelName: MODEL_NAME
      criticality: CRITICALITY
      poolRef:
        name: INFERENCE_POOL_NAME
    

    更改下列內容:

    • MODEL_NAME:基礎模型或 LoRA 轉接程式的名稱。例如:food-review
    • CRITICALITY:所選的放送重要性。選擇 CriticalStandardSheddable。例如:Standard
    • INFERENCE_POOL_NAME:您在上一個步驟中建立的InferencePool名稱。例如:vllm-llama3-8b-instruct
  2. 將範例資訊清單套用至叢集:

    kubectl apply -f inferencemodel.yaml
    

以下範例會建立 InferenceModel 物件,在 vllm-llama3-8b-instruct InferencePool 上設定 food-review LoRA 模型,並使用 Standard 服務重要性。InferenceModel 物件也會設定要以 Critical 優先順序層級放送的基礎模型。

apiVersion: inference.networking.x-k8s.io/v1alpha2
kind: InferenceModel
metadata:
  name: food-review
spec:
  modelName: food-review
  criticality: Standard
  poolRef:
    name: vllm-llama3-8b-instruct
  targetModels:
  - name: food-review
    weight: 100

---
apiVersion: inference.networking.x-k8s.io/v1alpha2
kind: InferenceModel
metadata:
  name: llama3-base-model
spec:
  modelName: meta-llama/Llama-3.1-8B-Instruct
  criticality: Critical
  poolRef:
    name: vllm-llama3-8b-instruct

建立閘道

Gateway 資源可做為外部流量進入 Kubernetes 叢集的進入點。定義接受連入連線的接聽程式。

GKE 推論閘道支援 gke-l7-rilbgke-l7-regional-external-managed 閘道類別。詳情請參閱 GKE 說明文件中的 Gateway 類別

如要建立閘道,請執行下列步驟:

  1. 將下列範例資訊清單儲存為 gateway.yaml

    apiVersion: gateway.networking.k8s.io/v1
    kind: Gateway
    metadata:
      name: GATEWAY_NAME
    spec:
      gatewayClassName: gke-l7-regional-external-managed
      listeners:
        - protocol: HTTP # Or HTTPS for production
          port: 80 # Or 443 for HTTPS
          name: http
    

    GATEWAY_NAME 替換成 Gateway 資源的專屬名稱。例如:inference-gateway

  2. 將資訊清單套用至叢集:

    kubectl apply -f gateway.yaml
    

建立 HTTPRoute 資源

在本節中,您將建立 HTTPRoute 資源,定義 Gateway 如何將傳入的 HTTP 要求轉送至 InferencePool

HTTPRoute 資源定義 GKE 閘道如何將傳入的 HTTP 要求轉送至後端服務 (即 InferencePool)。指定相符規則 (例如標頭或路徑),以及流量應轉送至的後端。

如要建立 HTTPRoute,請執行下列步驟:

  1. 將下列範例資訊清單儲存為 httproute.yaml

    apiVersion: gateway.networking.k8s.io/v1
    kind: HTTPRoute
    metadata:
      name: HTTPROUTE_NAME
    spec:
      parentRefs:
      - name: GATEWAY_NAME
      rules:
      - matches:
        - path:
            type: PathPrefix
            value: PATH_PREFIX
        backendRefs:
        - name: INFERENCE_POOL_NAME
          group: inference.networking.x-k8s.io
          kind: InferencePool
    

    更改下列內容:

    • HTTPROUTE_NAMEHTTPRoute資源的專屬名稱。例如:my-route
    • GATEWAY_NAME:您建立的 Gateway 資源名稱。例如:inference-gateway
    • PATH_PREFIX:用於比對傳入要求的路徑前置字元。例如,/ 可比對所有項目。
    • INFERENCE_POOL_NAME:要將流量導向的InferencePool資源名稱。例如:vllm-llama3-8b-instruct
  2. 將資訊清單套用至叢集:

    kubectl apply -f httproute.yaml
    

傳送推論要求

設定 GKE 推論閘道後,您就可以將推論要求傳送至已部署的模型。

如要傳送推論要求,請執行下列步驟:

  • 擷取閘道端點。
  • 建構格式正確的 JSON 要求。
  • 使用 curl 將要求傳送至 /v1/completions 端點。

根據輸入提示和指定參數生成文字。

  1. 如要取得 Gateway 端點,請執行下列指令:

    IP=$(kubectl get gateway/GATEWAY_NAME -o jsonpath='{.status.addresses[0].value}')
    PORT=PORT_NUMBER # Use 443 for HTTPS, or 80 for HTTP
    

    更改下列內容:

    • GATEWAY_NAME:Gateway 資源的名稱。
    • PORT_NUMBER:您在 Gateway 中設定的通訊埠號碼。
  2. 如要使用 curl 將要求傳送至 /v1/completions 端點,請執行下列指令:

    curl -i -X POST https://${IP}:${PORT}/v1/completions \
    -H 'Content-Type: application/json' \
    -H 'Authorization: Bearer $(gcloud auth print-access-token)' \
    -d '{
        "model": "MODEL_NAME",
        "prompt": "PROMPT_TEXT",
        "max_tokens": MAX_TOKENS,
        "temperature": "TEMPERATURE"
    }'
    

    更改下列內容:

    • MODEL_NAME:要使用的模型或 LoRA 轉接器名稱。
    • PROMPT_TEXT:模型的輸入提示。
    • MAX_TOKENS:回覆中生成的權杖數量上限。
    • TEMPERATURE:控制輸出內容的隨機性。如要取得確定性輸出內容,請使用 0 值;如要取得更多創意輸出內容,請使用較大的數字。

請注意下列行為:

  • 要求主體:要求主體可以包含其他參數,例如 stoptop_p。如需完整的選項清單,請參閱 OpenAI API 規格
  • 錯誤處理:在用戶端程式碼中實作適當的錯誤處理機制,處理回應中可能發生的錯誤。舉例來說,請檢查 curl 回應中的 HTTP 狀態碼。一般來說,如果狀態碼不是 200,就表示發生錯誤。
  • 驗證和授權:針對實際工作環境部署作業,請使用驗證和授權機制保護 API 端點。在要求中加入適當的標頭 (例如 Authorization)。

設定 Inference Gateway 的可觀測性

GKE 推論閘道可讓您觀測推論工作負載的健康狀態、效能和行為。這有助於找出及解決問題、提升資源使用效率,並確保應用程式的可靠性。您可以在 Cloud Monitoring 中透過 Metrics Explorer 查看這些可觀測性指標。

如要設定 GKE Inference Gateway 的觀測功能,請參閱「設定觀測功能」。

刪除已部署的資源

如要避免系統向您的 Google Cloud 帳戶收取您透過本指南建立的資源費用,請執行下列指令:

gcloud container clusters delete CLUSTER_NAME \
    --region=REGION

替換下列值:

  • REGION:支援您要使用的加速器類型,例如 H100 GPU 的 us-central1
  • CLUSTER_NAME:叢集名稱。

後續步驟