Implement runtime mode tracking and retry mechanism for Zero GPU initialization, falling back to CPU mode if necessary

app.py

@@ -39,6 +39,9 @@ except Exception as e:
     print(f"Zero GPU initialization warning: {e}")
     print("Running without Zero GPU support")
 
+# Runtime mode tracking
+RUNTIME_MODE = "GPU" if SPACES_AVAILABLE else "CPU"
+
 # Keep-alive state
 last_activity = datetime.datetime.now()
 activity_lock = threading.Lock()
@@ -437,10 +440,14 @@ with tranception_design:
     with gr.Row():
         with gr.Column(scale=1):
             def get_gpu_status():
+                global RUNTIME_MODE
                 with activity_lock:
                     time_since = (datetime.datetime.now() - last_activity).total_seconds()
                 
-                status = "🔥 Zero GPU" if SPACES_AVAILABLE else "💻 CPU Mode"
+                if RUNTIME_MODE == "GPU":
+                    status = "🔥 Zero GPU"
+                else:
+                    status = "💻 CPU Mode (GPU initialization failed)"
                 return f"{status} | Last activity: {int(time_since)}s ago"
             
             gpu_status = gr.Textbox(
@@ -546,10 +553,42 @@ if __name__ == "__main__":
     print("Note: Models will be downloaded on first use")
     print("Zero GPU spaces may sleep after ~15 minutes of inactivity")
     
-    # Launch with queue for proper Zero GPU support
-    tranception_design.queue(max_size=20).launch(
-        server_name="0.0.0.0",
-        server_port=7860,
-        show_error=True,
-        share=False
-    )
+    # Try to launch with ZeroGPU support first
+    launch_success = False
+    max_retries = 3
+    retry_count = 0
+    
+    while not launch_success and retry_count < max_retries:
+        try:
+            if retry_count > 0:
+                print(f"Retry attempt {retry_count}/{max_retries}...")
+                time.sleep(2)  # Wait before retry
+            
+            # Launch with queue for proper Zero GPU support
+            tranception_design.queue(max_size=20).launch(
+                server_name="0.0.0.0",
+                server_port=7860,
+                show_error=True,
+                share=False
+            )
+            launch_success = True
+        except RuntimeError as e:
+            if "Error while initializing ZeroGPU" in str(e):
+                retry_count += 1
+                if retry_count >= max_retries:
+                    print(f"ZeroGPU initialization failed after {max_retries} attempts")
+                    print("Falling back to CPU mode for stability")
+                    print("Note: The app will run slower in CPU mode")
+                    # Update runtime mode
+                    RUNTIME_MODE = "CPU"
+                    # Launch without queue which doesn't trigger ZeroGPU initialization
+                    tranception_design.launch(
+                        server_name="0.0.0.0",
+                        server_port=7860,
+                        show_error=True,
+                        share=False
+                    )
+                    launch_success = True
+            else:
+                # Re-raise unexpected errors
+                raise