fix: ZeroGPU server binding 0.0.0.0:7860

app.py

@@ -1,35 +1,84 @@
 """
 Alture AI — Hugging Face ZeroGPU Production Backend Entrypoint
 ============================================================
-FastAPI REST API Server mounted with Gradio ZeroGPU pipeline handler at /gradio.
+Official Hugging Face ZeroGPU production runner using user's proven monkeypatching pattern.
 """
 
 import os
-import uvicorn
+import sys
+import torch
 import gradio as gr
+from fastapi.middleware.cors import CORSMiddleware
+from fastapi.responses import Response
 
 try:
     import spaces
     @spaces.GPU
-    def zero_gpu_pipeline(resume_text: str):
-        return "ZeroGPU Pipeline Active & Ready"
+    def dummy_gpu_function():
+        return "ZeroGPU is active and ready"
 except ImportError:
-    def zero_gpu_pipeline(resume_text: str):
-        return "CPU Pipeline Active & Ready"
+    def dummy_gpu_function():
+        return "Local CPU mode"
 
-from deployment.backend.main import app as fastapi_app
+from deployment.backend.main import app as main_fastapi_app
 
-# Create Gradio demo to satisfy ZeroGPU SDK requirement
+# Create Gradio demo to satisfy ZeroGPU compiler
 with gr.Blocks(title="Alture AI Backend API") as demo:
     gr.Markdown("# Alture AI — Production ZeroGPU Backend Engine")
-    gr.Markdown("FastAPI REST endpoints available at `/api/v1` for Vercel Frontend integration.")
+    gr.Markdown("FastAPI REST endpoints available under `/api/v1` and `/v1`.")
     btn = gr.Button("⚡ Verify ZeroGPU Pipeline", variant="primary")
     out = gr.Textbox(label="Pipeline Output")
-    btn.click(zero_gpu_pipeline, inputs=gr.Textbox(value="Sample Resume"), outputs=out)
+    btn.click(dummy_gpu_function, outputs=out)
 
-# Mount Gradio onto FastAPI app at /gradio so root FastAPI routes remain 100% clean
-app = gr.mount_gradio_app(fastapi_app, demo, path="/gradio")
+# Monkeypatch Gradio's internal FastAPI app creator (proven portfolio pattern)
+original_create_app = gr.routes.App.create_app
+
+def custom_create_app(*args, **kwargs):
+    app = original_create_app(*args, **kwargs)
+    
+    # Configure CORS & preflight middleware
+    app.add_middleware(
+        CORSMiddleware,
+        allow_origins=["*"],
+        allow_credentials=True,
+        allow_methods=["*"],
+        allow_headers=["*"],
+    )
+
+    @app.middleware("http")
+    async def cors_preflight_middleware(request, call_next):
+        if request.method == "OPTIONS":
+            return Response(
+                status_code=200,
+                headers={
+                    "Access-Control-Allow-Origin": "*",
+                    "Access-Control-Allow-Methods": "*",
+                    "Access-Control-Allow-Headers": "*",
+                }
+            )
+        response = await call_next(request)
+        response.headers["Access-Control-Allow-Origin"] = "*"
+        response.headers["Access-Control-Allow-Methods"] = "*"
+        response.headers["Access-Control-Allow-Headers"] = "*"
+        return response
+
+    # Inject main FastAPI app router
+    app.include_router(main_fastapi_app.router)
+    
+    # Reorder routes so /api and /v1 routes take precedence
+    api_routes = []
+    other_routes = []
+    for route in app.router.routes:
+        path = getattr(route, 'path', '')
+        if path.startswith("/api") or path.startswith("/v1") or path.startswith("/health"):
+            api_routes.append(route)
+        else:
+            other_routes.append(route)
+            
+    app.router.routes = api_routes + other_routes
+    return app
+
+gr.routes.App.create_app = custom_create_app
 
 if __name__ == "__main__":
-    print("🚀 Launching Alture AI ZeroGPU Engine via demo.launch()...")
-    demo.launch()
+    demo.launch(server_name="0.0.0.0", server_port=7860)