Update app.py

app.py

@@ -1,8 +1,7 @@
 import spaces  # must be first, before torch/transformers
 import gradio as gr
-import uvicorn
-from fastapi import FastAPI
 from fastapi.middleware.cors import CORSMiddleware
+from starlette.routing import Mount
 
 from app.main import app as moderation_api, moderate_text, TextRequest, generate_suggestions
 
@@ -35,24 +34,28 @@ with gr.Blocks(title="Multilingual Moderation API") as demo:
     out = gr.Textbox(lines=12, label="Moderation Result")
     btn.click(moderate_ui, inp, out, api_name="moderate_ui")
 
-    # Hidden event so ZeroGPU detects the @spaces.GPU function at startup
+    # Hidden event so the @spaces.GPU function is registered with Gradio
     with gr.Row(visible=False):
         h_in = gr.Textbox()
         h_out = gr.JSON()
         h_btn = gr.Button()
         h_btn.click(generate_suggestions, h_in, h_out, api_name="rewrite")
 
-
-root = FastAPI()
-root.add_middleware(
+# CORS goes on the sub-app (middleware can't be added to Gradio's app after launch)
+moderation_api.add_middleware(
     CORSMiddleware,
     allow_origins=["*"],  # restrict to your Angular domain in production
     allow_methods=["*"],
     allow_headers=["*"],
 )
 
-root.mount("/api", moderation_api)  # -> /api/moderate and /api/health
-app = gr.mount_gradio_app(root, demo, path="/")
-
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=7860)
+    # launch() is what triggers ZeroGPU's startup detection
+    gradio_app, _, _ = demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        prevent_thread_lock=True,
+    )
+    # Put /api first so it wins over any Gradio route
+    gradio_app.router.routes.insert(0, Mount("/api", app=moderation_api))
+    demo.block_thread()