force backend

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

@@ -47,6 +47,8 @@ metadata:
   namespace: nim-service
 spec:
   env:
+    - 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:
@@ -59,7 +61,6 @@ spec:
   storage:
     nimCache:
       name: qwen36-27b-fp8
-      profile: vllm
   replicas: 1
   resources:
     limits: