diff --git a/clusters/k3s-dgx/nim-service/qwen.yaml b/clusters/k3s-dgx/nim-service/qwen.yaml
index 0f1b79f..872b8e9 100644
--- a/clusters/k3s-dgx/nim-service/qwen.yaml
+++ b/clusters/k3s-dgx/nim-service/qwen.yaml
@@ -64,10 +64,6 @@ spec:
     - --max-num-seqs
     - "254"                   # ↑ Allow more concurrent sequences
     - --enable-prefix-caching
-    - --dtype
-    - "float"                # Explicit FP8 encoding
-    - --quantization
-    - "fp8"            # Explicit quantization backend
   authSecret: hf-api-secret
   storage:
     sharedMemorySizeLimit: 64Gi