Fix: Add API routes after queue setup, avoid interfering with Gradio static assets

app.py

@@ -463,14 +463,18 @@ with gr.Blocks(
         outputs=[prompt_input, output],
     )
     
-    # Mount FastAPI routes directly onto Gradio's app (inside Blocks context)
-    # Add routes immediately so they're registered before Gradio processes requests
+    # Note: API routes will be added after Blocks context to avoid interfering with Gradio's static assets
+
+# Add API routes after Blocks context and queue setup
+# This ensures Gradio's static assets are fully initialized first
+def add_api_routes_safely():
+    """Add API routes without interfering with Gradio's static asset serving."""
     try:
         from fastapi.responses import JSONResponse
         from starlette.routing import Route
         from starlette.requests import Request
         
-        # Create route handlers that call the FastAPI endpoint functions
+        # Create route handlers
         async def health_handler(request: Request):
             """Handle GET /health requests."""
             result = healthcheck()
@@ -481,44 +485,24 @@ with gr.Blocks(
             try:
                 data = await request.json()
                 payload = GeneratePayload(**data)
-                # Call the FastAPI endpoint function directly
                 result = generate_endpoint(payload)
                 return JSONResponse(content={"text": result.text})
             except Exception as exc:
                 from fastapi import HTTPException
                 raise HTTPException(status_code=500, detail=str(exc))
         
-        # Store handlers to add routes after Blocks context (router is ready then)
-        # Store in a list that will be accessible after the Blocks context
-        if not hasattr(gradio_app, '_api_handlers'):
-            gradio_app._api_handlers = []
-        gradio_app._api_handlers.extend([
-            ("/health", health_handler, ["GET"]),
-            ("/v1/generate", generate_handler, ["POST"]),
-        ])
+        # Add routes directly - Starlette will match them in order
+        # Gradio's routes (like /_app/*) should already be registered
+        gradio_app.app.router.routes.append(Route("/health", health_handler, methods=["GET"]))
+        gradio_app.app.router.routes.append(Route("/v1/generate", generate_handler, methods=["POST"]))
+        print(f"API routes added. Total routes: {len(gradio_app.app.router.routes)}")
     except Exception as e:
-        print(f"Warning: Could not mount FastAPI routes: {e}")
+        print(f"Warning: Could not add API routes: {e}")
         import traceback
         traceback.print_exc()
 
-# Enable queued execution so ZeroGPU can schedule GPU work reliably
-gradio_app.queue(max_size=8)
-
-# Add routes after Blocks context when router is definitely ready
-# Append routes instead of inserting to avoid interfering with Gradio's static asset routes
-try:
-    if hasattr(gradio_app, '_api_handlers'):
-        from starlette.routing import Route
-        for path, handler, methods in gradio_app._api_handlers:
-            # Append routes so Gradio's routes (especially /_app/*) are checked first
-            gradio_app.app.router.routes.append(Route(path, handler, methods=methods))
-        print(f"FastAPI routes added after Blocks context. Total routes: {len(gradio_app.app.router.routes)}")
-        # Clean up
-        del gradio_app._api_handlers
-except Exception as e:
-    print(f"Warning: Could not add routes after Blocks context: {e}")
-    import traceback
-    traceback.print_exc()
+# Call after queue setup
+add_api_routes_safely()
 
 # Set app to Gradio for Spaces compatibility (sdk: gradio requires Gradio app)
 # Spaces will handle running the server automatically