Update app.py

app.py

@@ -24,7 +24,6 @@ from typing import Any, Dict, Iterator, List, Optional, Tuple, Union
 
 import gradio as gr
 import torch
-import uvicorn
 from fastapi import Depends, FastAPI, Header, HTTPException, Request
 from fastapi.responses import JSONResponse, StreamingResponse
 from pydantic import BaseModel, ConfigDict
@@ -466,8 +465,23 @@ with gr.Blocks(title="Myosotis API") as demo:
         api_name=False,  # keep Gradio's own API surface out of the way
     )
 
-# API routes were registered first, so they win over the Gradio mount at "/".
-app = gr.mount_gradio_app(api, demo, path="/")
+def attach_api_to_gradio():
+    """Copy the /v1 routes onto Gradio's own FastAPI app (same server, same port)."""
+    gradio_app = demo.app
+    api_routes = [r for r in api.routes if getattr(r, "path", "").startswith(("/v1", "/health"))]
+    # Insert at the front so no Gradio catch-all route can shadow them.
+    for route in reversed(api_routes):
+        gradio_app.router.routes.insert(0, route)
+    gradio_app.add_exception_handler(HTTPException, openai_style_errors)
+
 
 if __name__ == "__main__":
-    uvicorn.run(app, host="0.0.0.0", port=int(os.getenv("PORT", "7860")))
+    # Let Gradio own the port (Spaces expects it), then add the OpenAI routes.
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=int(os.getenv("PORT", "7860")),
+        prevent_thread_lock=True,
+        ssr_mode=False,
+    )
+    attach_api_to_gradio()
+    demo.block_thread()