perf: Run on CPU to avoid ZeroGPU duration limits

app.py

@@ -85,7 +85,8 @@ collection = _default_client.get_or_create_collection(
 
 
 def get_device():
-    return "cuda" if torch.cuda.is_available() else "cpu"
+    # Use CPU to avoid ZeroGPU duration limits
+    return "cpu"
 
 
 def get_whisper_model():
@@ -524,24 +525,14 @@ def _process_youtube_impl(
         return f"Error analyzing video: {error_msg}"
 
 
-# Apply ZeroGPU decorator if available
-if ZEROGPU_AVAILABLE:
-    @spaces.GPU(duration=120)  # 2 minute timeout for video processing
-    def process_youtube(
-        url: str,
-        num_frames: int,
-        session_state: SessionState | None = None,
-        progress: gr.Progress = gr.Progress(),
-    ) -> str:
-        return _process_youtube_impl(url, num_frames, session_state, progress)
-else:
-    def process_youtube(
-        url: str,
-        num_frames: int,
-        session_state: SessionState | None = None,
-        progress: gr.Progress = gr.Progress(),
-    ) -> str:
-        return _process_youtube_impl(url, num_frames, session_state, progress)
+# Run on CPU to avoid ZeroGPU duration limits
+def process_youtube(
+    url: str,
+    num_frames: int,
+    session_state: SessionState | None = None,
+    progress: gr.Progress = gr.Progress(),
+) -> str:
+    return _process_youtube_impl(url, num_frames, session_state, progress)
 
 
 def chat_with_videos(