Ollama + Open WebUI 使用随记

Ollama 是简化本地设备上大型语言模型 (LLM) 安装和管理的平台，而 Open WebUI 是 Ollama LLM 运行程序的可扩展网络用户界面。

Ollama

在 Kubernetes 上部署 Ollama：

helm repo add ollama-helm https://otwld.github.io/ollama-helm/
helm repo update

cat <<EOF > ollama-values.yaml
replicaCount: 1
knative:
  enabled: false
image:
  repository: ollama/ollama
  pullPolicy: IfNotPresent
ollama:
  gpu:
    enabled: true
    type: 'nvidia'
    number: 1
    nvidiaResource: "nvidia.com/gpu"
    mig:
      enabled: false
  models:
    pull: []
    run: []
    create: []
  insecure: false
serviceAccount:
  create: true
  automount: true
runtimeClassName: "nvidia"
service:
  type: ClusterIP
  port: 11434
  nodePort: 31434
  loadBalancerIP:
ingress:
  enabled: false
livenessProbe:
  enabled: true
  path: /
  initialDelaySeconds: 60
  periodSeconds: 10
  timeoutSeconds: 5
  failureThreshold: 6
  successThreshold: 1
readinessProbe:
  enabled: true
  path: /
  initialDelaySeconds: 30
  periodSeconds: 5
  timeoutSeconds: 3
  failureThreshold: 6
  successThreshold: 1
autoscaling:
  enabled: false
persistentVolume:
  enabled: true
  accessModes:
    - ReadWriteOnce
  size: 10Gi
  storageClass: "longhorn"
updateStrategy:
  type: "Recreate"
hostIPC: false
hostPID: false
hostNetwork: false
EOF

helm -n ollama upgrade --install ollama ollama-helm/ollama -f ollama-values.yaml --version 1.7.0

下载 deepseek-r1:1.5b 模型并运行：

调用 API 运行：

root@rke2-cilium-01:~/ai-related# curl -s -X POST http://<ollama-svc-ip>:11434/api/generate -d '{ "model": "deepseek-r1:1.5b", "prompt": "你好?", "stream": false }' | jq
{
  "model": "deepseek-r1:1.5b",
  "created_at": "2025-02-27T04:46:43.19616833Z",
  "response": "<think>\n\n</think>\n\n您好！很高兴为您服务。请问有什么可以帮助您的？",
  "done": true,
  "done_reason": "stop",
  "context": [
    151644,
    108386,
    30,
    151645,
    151648,
    271,
    151649,
    271,
    111308,
    6313,
    112169,
    102804,
    47874,
    1773,
    109194,
    104139,
    111728,
    101214,
    11319
  ],
  "total_duration": 3243229672,
  "load_duration": 2905513629,
  "prompt_eval_count": 5,
  "prompt_eval_duration": 82000000,
  "eval_count": 16,
  "eval_duration": 253000000
}

Open WebUI

在 Kubernetes 上部署 Open WebUI：

helm repo add open-webui https://helm.openwebui.com/
helm repo update

cat <<EOF > open-webui-values.yaml
ollama:
  enabled: false
  fullnameOverride: "open-webui-ollama"
pipelines:
  enabled: false
tika:
  enabled: false
ollamaUrls:
  - http://ollama.ollama.svc.cluster.local:11434
ollamaUrlsFromExtraEnv: false
websocket:
  enabled: false
redis-cluster:
  enabled: false
clusterDomain: cluster.local
replicaCount: 1
image:
  repository: harbor.warnerchen.com/open-webui/open-webui
  pullPolicy: "IfNotPresent"
serviceAccount:
  enable: true
  automountServiceAccountToken: false
managedCertificate:
  enabled: false
ingress:
  enabled: true
  class: "nginx"
  tls: false
  host: "open-webui.warnerchen.com"
persistence:
  enabled: true
  size: 2Gi
  accessModes:
    - ReadWriteOnce
  storageClass: "longhorn"
service:
  type: ClusterIP
  port: 80
  containerPort: 8080
openaiBaseApiUrl: "https://api.openai.com/v1"
extraEnvVars:
  - name: OPENAI_API_KEY
    value: "0p3n-w3bu!"
  - name: WEBUI_AUTH
    value: "False"
EOF

helm -n ollama upgrade --install open-webui open-webui/open-webui -f open-webui-values.yaml --version 5.20.0

访问 Open WebUI：