Feature(MInference): install pycuda

app.py

@@ -81,12 +81,12 @@ def chat_llama3_8b(message: str,
         str: The generated response.
     """
     global model
+    subprocess.run(
+        "pip install pycuda==2023.1",
+        shell=True,
+    )
     if "has_patch" not in model.__dict__:
         from minference import MInference
-        subprocess.run(
-            "pip install pycuda==2023.1",
-            shell=True,
-        )
         minference_patch = MInference("minference", model_name)
         model = minference_patch(model)
     conversation = []