set VLLM_USE_NN=1

vllm/envs.py

@@ -969,7 +969,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
     
     # If set, vLLM will transpose weight to use nn layout
     "VLLM_USE_NN":
-    lambda: (os.environ.get("VLLM_USE_NN", "False").lower() in
+    lambda: (os.environ.get("VLLM_USE_NN", "True").lower() in
              ("true", "1")),
 
     # Enable two batch overlap.