Disable SSR and suppress spaces runtime warning

app.py

@@ -1,9 +1,16 @@
 import os
+import warnings
 
 from inference.interface import PenguinVLQwen3GradioInterface
 from inference.server import PenguinVLQwen3DirectClient
 from inference.server.direct_client import ensure_flash_attn_installed, preload_model
 
+warnings.filterwarnings(
+    "ignore",
+    message=r"`torch\.distributed\.reduce_op` is deprecated, please use `torch\.distributed\.ReduceOp` instead",
+    category=FutureWarning,
+)
+
 
 def main():
     model_path = os.getenv("MODEL_PATH", "tencent/Penguin-VL-8B")

inference/interface/gradio_interface.py

@@ -232,4 +232,9 @@ class PenguinVLQwen3GradioInterface(object):
             yield messages, ""
 
     def launch(self):
-        self.interface.launch(**self.server_kwargs)
+        launch_kwargs = dict(self.server_kwargs)
+        launch_kwargs.setdefault(
+            "ssr_mode",
+            os.getenv("PENGUINVL_GRADIO_SSR_MODE", "").strip().lower() in {"1", "true", "yes", "on"},
+        )
+        self.interface.launch(**launch_kwargs)