change backend

clusters/k3s-dgx/nim-service/qwen.yaml

@@ -46,26 +46,30 @@ metadata:
   name: qwen36-27b-fp8
   namespace: nim-service
 spec:
-  env:
-    - name: NIM_DISABLE_MODEL_DOWNLOAD
-      value: "1"
-    - name: HF_HUB_OFFLINE
-      value: "1"
-    - name: NIM_BACKEND
-      value: vllm
-    - name: NIM_PASSTHROUGH_ARGS
-      value: "--gpu-memory-utilization 0.85 --served-model-name qwen36 --max-model-len 256K --language-model-only --reasoning-parser qwen3 --enable-auto-tool-choice --tool-call-parser qwen3_coder --enable-chunked-prefill --max-num-batched-tokens 32768 --max-num-seqs 10 --enable-prefix-caching --speculative-config '{\"method\":\"mtp\",\"num_speculative_tokens\":2}'"
   image:
-    repository: nvcr.io/nim/nvidia/llm-nim
-    tag: "1.12"
+    repository: vllm/vllm-openai
+    tag: latest
     pullPolicy: IfNotPresent
-    pullSecrets:
-      - ngc-secret
+
+  command:
+    - python3
+
+  args:
+    - -m
+    - vllm.entrypoints.openai.api_server
+    - --model
+    - /model-store
+    - --served-model-name
+    - qwen36
+    - --gpu-memory-utilization
+    - "0.85"
+    - --max-model-len
+    - "262144"
+    - --enable-prefix-caching
   authSecret: hf-api-secret
   storage:
     nimCache:
       name: qwen36-27b-fp8
-      profile: Qwen3_5ForConditionalGeneration
   replicas: 1
   resources:
     limits: