Pass VLLM_USE_OUTLINES=0 to subprocess and remove outlines dependencies

app.py

@@ -34,7 +34,9 @@ if "AWQ" in MODEL_ID.upper():
 
 def launch_vllm():
     print(f"[vLLM] Launch: {MODEL_ID}")
-    subprocess.Popen(VLLM_ARGS)
+    env = os.environ.copy()
+    env["VLLM_USE_OUTLINES"] = "0"  # disable outlines to avoid pyairports import
+    subprocess.Popen(VLLM_ARGS, env=env)
 
 def wait_vllm_ready(timeout=900, interval=3):
     url = f"http://127.0.0.1:{API_PORT}/v1/models"

requirements.txt

@@ -10,8 +10,8 @@ torch==2.4.0
 transformers>=4.44
 accelerate>=0.30
 
-# Structured outputs stack used by vLLM
-outlines[types]==0.0.46
-numba==0.59.1
-pycountry==24.6.1
-pyairports==0.0.1
+# Structured outputs stack used by vLLM (disabled via VLLM_USE_OUTLINES=0)
+# outlines[types]==0.0.46
+# numba==0.59.1
+# pycountry==24.6.1
+# pyairports==0.0.1