udpate vllm

charts/vllm/custom-values.yaml

@@ -1,27 +1,34 @@
 servingEngineSpec:
   modelSpec:
-  - name: "llama3"
+    - name: "qwen"
       repository: "vllm/vllm-openai"
       tag: "latest"
-    modelURL: "meta-llama/Llama-3.1-8B-Instruct"
+      modelURL: "Qwen/Qwen2.5-Coder-14B-Instruct"
       replicaCount: 1
 
-    requestCPU: 10
-    requestMemory: "16Gi"
+      requestCPU: 16
+      requestMemory: "32Gi"
       requestGPU: 1
 
-    pvcStorage: "50Gi"
+      pvcStorage: "10Gi"
+      pvcAccessMode:
+        - ReadWriteMany
+      storageClass: "nfs"
 
       vllmConfig:
         enableChunkedPrefill: false
         enablePrefixCaching: false
-      maxModelLen: 24576
+        maxModelLen: 4096
         dtype: "float16"
-      extraArgs: ["--disable-log-requests", "--gpu-memory-utilization", "0.8"]
-    hf_token: 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
-
+        extraArgs:
+          [
+            "--disable-log-requests",
+            "--gpu-memory-utilization",
+            "0.95",
+            "--download-dir",
+            "/data/qwen",
+          ]
 routerSpec:
-  # -- The docker image of the router. The following values are defaults:
   repository: "lmcache/lmstack-router"
   tag: "latest"
   enableRouter: true