Fix: Use gradio_app.load() event to add routes after initialization

app.py

@@ -511,54 +511,63 @@ with gr.Blocks(
         outputs=[prompt_input, output],
     )
     
-    # Add FastAPI routes directly to Gradio's router
-    # This ensures they're registered and accessible alongside Gradio routes
-    try:
-        from fastapi.responses import JSONResponse
-        
-        async def generate_handler(request):
-            """Handle POST /v1/generate requests."""
-            try:
-                data = await request.json()
-                payload = GeneratePayload(**data)
-                text = _generate_with_gpu(
-                    prompt=payload.prompt,
-                    max_new_tokens=payload.max_new_tokens or MAX_NEW_TOKENS,
-                    temperature=payload.temperature or DEFAULT_TEMPERATURE,
-                    top_p=payload.top_p or DEFAULT_TOP_P,
-                )
-                return JSONResponse(content={"text": text})
-            except Exception as exc:
-                from fastapi import HTTPException
-                raise HTTPException(status_code=500, detail=str(exc))
-        
-        async def healthcheck_handler(request):
-            """Handle GET /api/health requests."""
-            return JSONResponse(content={
-                "status": "ok",
-                "model": MODEL_ID,
-                "strategy": ACTIVE_STRATEGY or "pending",
-            })
-        
-        async def gradio_ui_handler(request):
-            """Handle GET /api/gradio requests."""
-            return HTMLResponse(interactive_ui())
-        
-        # Add routes to Gradio's router using Starlette's router
-        # Use Starlette Route objects for compatibility
-        from starlette.routing import Route
-        
-        # Add routes to Gradio's router
-        gradio_app.app.router.add_route("/v1/generate", generate_handler, methods=["POST"])
-        gradio_app.app.router.add_route("/api/health", healthcheck_handler, methods=["GET"])
-        gradio_app.app.router.add_route("/api/gradio", gradio_ui_handler, methods=["GET"])
-        # Also add /gradio for backward compatibility
-        gradio_app.app.router.add_route("/gradio", gradio_ui_handler, methods=["GET"])
-        print("FastAPI routes added successfully to Gradio router")
-    except Exception as e:
-        print(f"Warning: Could not add FastAPI routes: {e}")
-        import traceback
-        traceback.print_exc()
+    # Add FastAPI routes using Gradio's load event
+    # This ensures routes are added after Gradio is fully initialized
+    def add_api_routes():
+        """Add API routes after Gradio app is loaded."""
+        try:
+            from fastapi.responses import JSONResponse
+            from starlette.routing import Route
+            
+            async def generate_handler(request):
+                """Handle POST /v1/generate requests."""
+                try:
+                    data = await request.json()
+                    payload = GeneratePayload(**data)
+                    text = _generate_with_gpu(
+                        prompt=payload.prompt,
+                        max_new_tokens=payload.max_new_tokens or MAX_NEW_TOKENS,
+                        temperature=payload.temperature or DEFAULT_TEMPERATURE,
+                        top_p=payload.top_p or DEFAULT_TOP_P,
+                    )
+                    return JSONResponse(content={"text": text})
+                except Exception as exc:
+                    from fastapi import HTTPException
+                    raise HTTPException(status_code=500, detail=str(exc))
+            
+            async def healthcheck_handler(request):
+                """Handle GET /api/health requests."""
+                return JSONResponse(content={
+                    "status": "ok",
+                    "model": MODEL_ID,
+                    "strategy": ACTIVE_STRATEGY or "pending",
+                })
+            
+            async def gradio_ui_handler(request):
+                """Handle GET /api/gradio requests."""
+                return HTMLResponse(interactive_ui())
+            
+            # Add routes using Route objects
+            gradio_app.app.router.routes.append(
+                Route("/v1/generate", generate_handler, methods=["POST"])
+            )
+            gradio_app.app.router.routes.append(
+                Route("/api/health", healthcheck_handler, methods=["GET"])
+            )
+            gradio_app.app.router.routes.append(
+                Route("/api/gradio", gradio_ui_handler, methods=["GET"])
+            )
+            gradio_app.app.router.routes.append(
+                Route("/gradio", gradio_ui_handler, methods=["GET"])
+            )
+            print("FastAPI routes added successfully via load event")
+        except Exception as e:
+            print(f"Warning: Could not add FastAPI routes: {e}")
+            import traceback
+            traceback.print_exc()
+    
+    # Use load event to add routes after app initialization
+    gradio_app.load(add_api_routes)
 
 # Set app to Gradio Blocks for Spaces - ZeroGPU requires Gradio SDK
 app = gradio_app