[Feat] Add support for external Hugging Face token secrets in Helm ch…

…art (#149) * [Feat] Add support for external Hugging Face token secrets in Helm chart - Update values.schema.json to allow HF token as a string or secret reference - Modify values.yaml comments to clarify HF token configuration options - Update deployment and secrets templates to handle both token types - Fix a minor comment in PVC template about default storage size * [Fix] Correct Helm chart secret reference in vLLM multi-model deployment Signed-off-by: Liad Drori <[email protected]>
vllm-project · Feb 25, 2025 · 0143de5 · 0143de5
1 parent c10a2c2
commit 0143de5
Show file tree

Hide file tree

Showing 5 changed files with 26 additions and 5 deletions.
diff --git a/helm/templates/deployment-vllm-multi.yaml b/helm/templates/deployment-vllm-multi.yaml
@@ -81,10 +81,17 @@ spec:
           {{- end}}
           {{- if $modelSpec.hf_token }}
           - name: HF_TOKEN
+            {{- if kindIs "string" $modelSpec.hf_token }}
             valueFrom:
               secretKeyRef:
                 name: {{ .Release.Name }}-secrets
                 key: hf_token_{{ $modelSpec.name }}
+            {{- else }}
+            valueFrom:
+              secretKeyRef:
+                name: {{ $modelSpec.hf_token.secretName }}
+                key: {{ $modelSpec.hf_token.secretKey }}
+            {{- end }}
           {{- end }}
           {{- with $modelSpec.env }}
           {{- toYaml . | nindent 10 }}

diff --git a/helm/templates/pvc.yaml b/helm/templates/pvc.yaml
@@ -15,7 +15,7 @@ spec:
     {{- end }}
   resources:
     requests:
-      storage: {{ $modelSpec.pvcStorage | default "20Gi" }}  # Default to 40Gi if not set
+      storage: {{ $modelSpec.pvcStorage | default "20Gi" }}  # Default to 20Gi if not set
   {{- if hasKey $modelSpec "storageClass" }}
   storageClassName: "{{ $modelSpec.storageClass }}"
   {{- end }}

diff --git a/helm/templates/secrets.yaml b/helm/templates/secrets.yaml
@@ -7,7 +7,7 @@ type: Opaque
 data:
   {{- range $modelSpec := .Values.servingEngineSpec.modelSpec }}
   {{- with $ -}}
-  {{-   if $modelSpec.hf_token }}
+  {{-   if and $modelSpec.hf_token (kindIs "string" $modelSpec.hf_token) }}
   hf_token_{{ $modelSpec.name }}: {{ $modelSpec.hf_token | b64enc | quote }}
   {{-   end }}
   {{- end }}

diff --git a/helm/values.schema.json b/helm/values.schema.json
@@ -54,7 +54,17 @@
                 "required": ["enabled", "cpuOffloadingBufferSize"]
               },
               "hf_token": {
-                  "type": "string"
+                "oneOf": [
+                  { "type": "string" },
+                  {
+                    "type": "object",
+                    "properties": {
+                      "secretName": { "type": "string" },
+                      "secretKey": { "type": "string" }
+                    },
+                    "required": ["secretName", "secretKey"]
+                  }
+                ]
               },
               "env": {
                 "type": "array",

diff --git a/helm/values.yaml b/helm/values.yaml
@@ -38,7 +38,11 @@ servingEngineSpec:
   #   - enabled: (optional, bool) Enable LMCache, e.g., true
   #   - cpuOffloadingBufferSize: (optional, string) The CPU offloading buffer size, e.g., "30"
   #
-  # - hf_token: (optional, string) the Huggingface tokens for this model
+  # - hf_token: (optional) Hugging Face token configuration. Can be either:
+  #   - A string containing the token directly (will be stored in a generated secret)
+  #   - An object referencing an existing secret:
+  #     secretName: "my-existing-secret"
+  #     secretKey: "hf-token-key"
   #
   # - env: (optional, list) The environment variables to set in the container, e.g., your HF_TOKEN
   #
@@ -75,7 +79,7 @@ servingEngineSpec:
   #     enabled: true
   #     cpuOffloadingBufferSize: "30"
   #
-  #   hf_token: <HUGGING_FACE_TOKEN>
+  #   hf_token: "hf_xxxxxxxxxxxxx"
   #
   #   nodeSelectorTerms:
   #     - matchExpressions: