Update app.py

app.py

@@ -7,8 +7,6 @@ This follows the inference example in README.md and uses the
 
 from __future__ import annotations
 
-import asyncio
-import atexit
 import os
 import sys
 from typing import Optional, List, Tuple
@@ -39,36 +37,6 @@ _MODEL: Optional[AutoModelForCausalLM] = None
 _PROCESSOR: Optional[AutoProcessor] = None
 
 
-def _close_default_event_loop() -> None:
-    """Close the default asyncio loop to avoid fd errors on interpreter exit."""
-    try:
-        loop = asyncio.get_event_loop()
-    except RuntimeError:
-        return
-
-    if loop.is_closed():
-        return
-
-    # If the loop is still running, ask it to stop; closing while running raises.
-    if loop.is_running():
-        loop.stop()
-
-    try:
-        loop.run_until_complete(loop.shutdown_asyncgens())
-    except RuntimeError:
-        # If the loop was already running in another context, just try closing.
-        pass
-
-    try:
-        loop.close()
-    except Exception:
-        # Best-effort cleanup; swallow any finalizer noise.
-        pass
-
-
-atexit.register(_close_default_event_loop)
-
-
 def load_model_and_processor() -> Tuple[AutoModelForCausalLM, AutoProcessor]:
     """Lazy-load the QTSplus model and processor."""
     global _MODEL, _PROCESSOR
@@ -103,8 +71,8 @@ def load_model_and_processor() -> Tuple[AutoModelForCausalLM, AutoProcessor]:
     return _MODEL, _PROCESSOR
 
 
-# Preload model and processor at import time (for faster first inference).
-load_model_and_processor()
+# # Preload model and processor at import time (for faster first inference).
+# load_model_and_processor()
 
 
 def build_messages(video: Optional[str], prompt: str) -> List[dict]:
@@ -231,9 +199,4 @@ with gr.Blocks() as demo:
 
 
 if __name__ == "__main__":
-    try:
-        demo.queue().launch()
-    finally:
-        # Ensure Gradio and asyncio clean up cleanly to prevent fd errors.
-        demo.close()
-        _close_default_event_loop()
+    demo.queue().launch()