changed to uvicorn setup for HF v3

main/main.py

@@ -57,7 +57,15 @@ app.include_router(router, prefix=api_prefix)
 @app.on_event("startup")
 async def startup_event():
     """Initialize async components on startup."""
+    # Initialize the router
     await init_router(api)
+    # Launch the inference worker
+    server.launch_inference_worker()
+
+@app.on_event("shutdown")
+async def shutdown_event():
+    """Cleanup on shutdown."""
+    server.stop_inference_worker()
 
 async def run_server():
     """Run the server directly (not through uvicorn)"""