[Bugfix] value file based accessMode (#108)

* fix: value file based accessMode * fix: add config to examples --------- Signed-off-by: BrianPark314 <[email protected]>
vllm-project · Feb 11, 2025 · b470098 · b470098
1 parent 8d4b05a
commit b470098
Show file tree

Hide file tree

Showing 9 changed files with 22 additions and 2 deletions.
diff --git a/.github/multiple-models.yaml b/.github/multiple-models.yaml
@@ -10,6 +10,8 @@ servingEngineSpec:
     requestMemory: "16Gi"
     requestGPU: 1
     pvcStorage: "10Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
 
   - name: "smol135m"
     repository: "vllm/vllm-openai"
@@ -20,3 +22,5 @@ servingEngineSpec:
     requestMemory: "16Gi"
     requestGPU: 1
     pvcStorage: "10Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
diff --git a/helm/templates/pvc.yaml b/helm/templates/pvc.yaml
@@ -6,8 +6,7 @@ metadata:
   name: "{{ .Release.Name }}-{{$modelSpec.name}}-storage-claim"
   namespace: {{ .Release.Namespace }}
 spec:
-  accessModes:
-    - ReadWriteOnce
+  accessModes: {{ toYaml $modelSpec.pvcAccessMode | nindent 4 }}
   resources:
     requests:
       storage: {{ $modelSpec.pvcStorage }}

diff --git a/helm/values.yaml b/helm/values.yaml
@@ -22,6 +22,7 @@ servingEngineSpec:
   # - requestGPU: (int) The number of GPUs requested for the model, e.g., 1
   #
   # - pvcStorage: (string) The amount of storage requested for the model, e.g., "50Gi"
+  # - pvcAccessMode: (list) The access mode policy for the mounted volume, e.g., ["ReadWriteOnce"]
   # - pvcMatchLabels: (optional, map) The labels to match the PVC, e.g., {model: "opt125m"}
   #
   # - vllmConfig: (optional, map) The configuration for the VLLM model, supported options are:
@@ -57,6 +58,8 @@ servingEngineSpec:
   #   requestGPU: 1
   #
   #   pvcStorage: "50Gi"
+  #   pvcAccessMode:
+  #     - ReadWriteOnce
   #   pvcMatchLabels:
   #     model: "mistral"
   #

diff --git a/tutorials/assets/values-01-2pods-minimal-example.yaml b/tutorials/assets/values-01-2pods-minimal-example.yaml
@@ -13,6 +13,8 @@ servingEngineSpec:
     requestGPU: 0.5
 
     pvcStorage: "10Gi"
+    pvcAccessMode:
+      - ReadWriteMany
 
     vllmConfig:
       maxModelLen: 1024

diff --git a/tutorials/assets/values-01-minimal-example.yaml b/tutorials/assets/values-01-minimal-example.yaml
@@ -13,3 +13,5 @@ servingEngineSpec:
     requestGPU: 1
 
     pvcStorage: "10Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
diff --git a/tutorials/assets/values-02-basic-config.yaml b/tutorials/assets/values-02-basic-config.yaml
@@ -12,6 +12,8 @@ servingEngineSpec:
     requestGPU: 1
 
     pvcStorage: "50Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
 
     vllmConfig:
       enableChunkedPrefill: false

diff --git a/tutorials/assets/values-03-match-pv.yaml b/tutorials/assets/values-03-match-pv.yaml
@@ -12,6 +12,8 @@ servingEngineSpec:
     requestGPU: 1
 
     pvcStorage: "50Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
     pvcMatchLabels:
       model: "llama3-pv"
 

diff --git a/tutorials/assets/values-04-multiple-models.yaml b/tutorials/assets/values-04-multiple-models.yaml
@@ -10,6 +10,8 @@ servingEngineSpec:
     requestMemory: "16Gi"
     requestGPU: 1
     pvcStorage: "50Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
     vllmConfig:
       maxModelLen: 4096
     hf_token: <YOUR HF TOKEN FOR LLAMA3.1>
@@ -23,6 +25,8 @@ servingEngineSpec:
     requestMemory: "16Gi"
     requestGPU: 1
     pvcStorage: "50Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
     vllmConfig:
       maxModelLen: 4096
     hf_token: <YOUR HF TOKEN FOR MISTRAL>
diff --git a/tutorials/assets/values-05-cpu-offloading.yaml b/tutorials/assets/values-05-cpu-offloading.yaml
@@ -10,6 +10,8 @@ servingEngineSpec:
     requestMemory: "40Gi"
     requestGPU: 1
     pvcStorage: "50Gi"
+    pvcAccessMode:
+      - ReadWriteOnce
     vllmConfig:
       enableChunkedPrefill: false
       enablePrefixCaching: false