Update app.py

app.py

@@ -1,25 +1,43 @@
-# 1. MUST be the very first import
+# 1. CRITICAL: Import spaces before anything else that might touch CUDA
 import spaces
 import gradio as gr
 from fastapi import FastAPI
-from app.main import app as fastapi_app
+from fastapi.middleware.cors import CORSMiddleware
 
-# 2. Dummy GPU function
+# Import your moderation router
+from app.api.v1.endpoints.moderation import router as moderation_router
+
+# 2. Dummy GPU function that MUST be wired to a Gradio event
 @spaces.GPU
 def dummy_gpu_check():
     return "ZeroGPU Hardware Connected!"
 
-# 3. Create the Gradio UI with a button that calls the GPU function
+# 3. Build the Gradio UI and wire the GPU function to a button click
 with gr.Blocks() as demo:
     gr.Markdown("# 🛡️ Moderation API")
     gr.Markdown("This is a live API. Send POST requests to `/api/v1/moderate`.")
     gr.Markdown("You can view the interactive API docs at [here](/docs) once it starts.")
     
-    # This button satisfies the ZeroGPU AST parser
     with gr.Row():
         btn = gr.Button("Test ZeroGPU Connection")
         output = gr.Textbox()
+        # The function passed here is the one decorated with @spaces.GPU
         btn.click(fn=dummy_gpu_check, inputs=[], outputs=output)
 
-# 4. Mount FastAPI
-app = gr.mount_gradio_app(fastapi_app, demo, path="/")
+# 4. Get Gradio's internal FastAPI app
+app = demo.app
+
+# 5. Attach your FastAPI routes and middleware to Gradio's app
+app.include_router(moderation_router, prefix="/api/v1")
+
+app.add_middleware(
+    CORSMiddleware,
+    allow_origins=["*"],
+    allow_credentials=True,
+    allow_methods=["*"],
+    allow_headers=["*"],
+)
+
+# 6. Launch Gradio (this initializes @spaces.GPU properly)
+if __name__ == "__main__":
+    demo.launch()