Update vllm to version 0.9.2 with enhanced model and configuration

vllm/versions/0/deployment.yaml

@@ -23,7 +23,7 @@ spec:
         feature.node.kubernetes.io/pci-0300_10de.present: "true"
       containers:
         - name: vllm
-          image: vllm/vllm-openai:v0.5.4
+          image: vllm/vllm-openai:v0.9.2
           imagePullPolicy: IfNotPresent
           securityContext:
             allowPrivilegeEscalation: false
@@ -36,6 +36,7 @@ spec:
             - --tensor-parallel-size=1
             - --gpu-memory-utilization={{ .gpuMemoryUtilization }}
             - --enforce-eager
+            - --reasoning-parser=deepseek_r1
             - --api-key=$(VLLM_API_KEY)
           env:
             - name: VLLM_API_KEY
@@ -51,6 +52,9 @@ spec:
               value: "/root/.cache/huggingface"
             - name: NUMBA_CACHE_DIR
               value: "/tmp"
+          volumeMounts:
+            - name: hf-cache
+              mountPath: /root/.cache/huggingface
           ports:
             - name: http
               containerPort: 8000
@@ -75,10 +79,13 @@ spec:
             httpGet:
               path: /health
               port: http
-            initialDelaySeconds: 600
+            initialDelaySeconds: 1800
             periodSeconds: 60
             timeoutSeconds: 10
             failureThreshold: 3
+      volumes:
+        - name: hf-cache
+          emptyDir: {}
       tolerations:
         - key: "nvidia.com/gpu"
           operator: "Exists"

vllm/versions/0/manifest.yaml

@@ -1,8 +1,8 @@
-version: 0.5.4-7
+version: 0.9.2-1
 defaultConfig:
   namespace: llm
-  model: Qwen/Qwen2.5-7B-Instruct
+  model: Qwen/Qwen3-8B
-  maxModelLen: 8192
+  maxModelLen: 32768
   gpuMemoryUtilization: 0.9
   cpuRequest: '4'
   cpuLimit: '8'