Use Gradio load event to mount FastAPI routes

app.py

@@ -411,31 +411,29 @@ with gr.Blocks(title="Router Model API - ZeroGPU") as gradio_app:
         inputs=[prompt_input, max_tokens_input, temp_input, top_p_input],
         outputs=output,
     )
-
-# Mount FastAPI routes to Gradio's underlying FastAPI app after Blocks context
-# Use a callback to mount routes when the app is ready
-def mount_fastapi_on_load():
-    """Mount FastAPI routes when Gradio app loads."""
-    try:
-        # Access Gradio's FastAPI app and mount our routes
-        gradio_app.app.mount("/v1", fastapi_app)
-        gradio_app.app.mount("/gradio", fastapi_app)
-    except (AttributeError, Exception) as e:
-        print(f"Note: FastAPI routes mounting deferred: {e}")
-
-# Try to mount immediately, and also set up for when app launches
-try:
-    mount_fastapi_on_load()
-except:
-    # If mounting fails now, it will be handled during launch
-    pass
-
-# Override launch to mount routes
-original_launch = gradio_app.launch
-def launch_with_mount(*args, **kwargs):
-    mount_fastapi_on_load()
-    return original_launch(*args, **kwargs)
-gradio_app.launch = launch_with_mount
+    
+    # Mount FastAPI routes using Gradio's load event
+    def mount_fastapi_routes():
+        """Mount FastAPI routes when Gradio app loads."""
+        try:
+            # Access Gradio's underlying FastAPI app
+            gradio_app.app.mount("/v1", fastapi_app)
+            gradio_app.app.mount("/gradio", fastapi_app)
+            print("FastAPI routes mounted successfully")
+        except (AttributeError, Exception) as e:
+            print(f"FastAPI routes mounting note: {e}")
+            # Try alternative mounting approach
+            try:
+                import starlette
+                # Mount using Starlette's mount
+                if hasattr(gradio_app, 'app') and hasattr(gradio_app.app, 'mount'):
+                    gradio_app.app.mount("/v1", fastapi_app)
+                    gradio_app.app.mount("/gradio", fastapi_app)
+            except Exception as e2:
+                print(f"Alternative mounting also failed: {e2}")
+    
+    # Use load event to mount routes when app is ready
+    gradio_app.load(mount_fastapi_routes)
 
 # Set app to Gradio Blocks for Spaces - ZeroGPU requires Gradio SDK
 app = gradio_app