fix: keep Gradio native launch; monkey-patch /video onto its own FastAPI (fixes /_app/immutable 404)

app.py

@@ -504,19 +504,19 @@ if __name__ == "__main__":
     # <video> sources (hf.space iframe → xet CDN). We dodge this by streaming
     # the mp4 through this Space's own FastAPI app so the browser sees the
     # video on the same origin as the page.
+    #
+    # IMPORTANT: we attach to Gradio's *own* FastAPI app (populated lazily
+    # during `.launch()`). Using gr.mount_gradio_app + a hand-built FastAPI
+    # was a mistake — it drops Gradio's `/_app/immutable/*` static-asset
+    # routes, causing the UI to render as a bare HTML skeleton on
+    # browsers that haven't cached the old bundle.
     import httpx
-    import uvicorn
-    from fastapi import FastAPI, HTTPException, Request
+    from fastapi import HTTPException, Request
     from fastapi.responses import StreamingResponse
+    from gradio.routes import App as _GradioApp
 
-    fastapi_app = FastAPI()
     _video_client = httpx.AsyncClient(timeout=30.0, follow_redirects=True)
 
-    @fastapi_app.api_route(
-        "/video/{model}/{task_id}.mp4",
-        methods=["GET", "HEAD"],
-        include_in_schema=False,
-    )
     async def _proxy_video(model: str, task_id: str, request: Request):
         if model not in MODELS or task_id not in TASK_BY_ID:
             raise HTTPException(404, "unknown (model, task_id)")
@@ -552,8 +552,23 @@ if __name__ == "__main__":
             headers=passthrough_headers,
         )
 
-    demo.queue(default_concurrency_limit=16)
-    fastapi_app = gr.mount_gradio_app(fastapi_app, demo, path="/")
-    print("[mbench-ann] mounted Gradio at '/' with /video proxy route")
-    uvicorn.run(fastapi_app, host="0.0.0.0", port=7860)
+    # Monkey-patch Gradio's App.create_app so that right after Gradio builds
+    # its own FastAPI app (with all the /_app/immutable/* routes, /gradio_api
+    # endpoints, etc.) we register the video proxy on the same instance.
+    _orig_create_app = _GradioApp.create_app
+
+    def _patched_create_app(*args, **kwargs):
+        app = _orig_create_app(*args, **kwargs)
+        app.add_api_route(
+            "/video/{model}/{task_id}.mp4",
+            _proxy_video,
+            methods=["GET", "HEAD"],
+            include_in_schema=False,
+        )
+        print("[mbench-ann] /video proxy route registered on Gradio's own app")
+        return app
+
+    _GradioApp.create_app = staticmethod(_patched_create_app)
+
+    demo.queue(default_concurrency_limit=16).launch()