add vllm

app.py

@@ -43,6 +43,7 @@ with open(f'{model_path}/tekken.json', 'r') as f:
 
 llm = None
 
+@spaces.GPU()
 def initialize_llm():
     global llm
     if llm is None: