Update app.py

app.py

@@ -1,7 +1,11 @@
 import gradio as gr
 import json
 
-def chat_completions(message):
+def chat_completions(request: gr.Request):
+    # Parse the incoming JSON request
+    data = request.json
+    
+    # Create a placeholder response
     response = {
         "id": "chatcmpl-123",
         "object": "chat.completion",
@@ -10,7 +14,7 @@ def chat_completions(message):
             "index": 0,
             "message": {
                 "role": "assistant",
-                "content": "This is a placeholder response. The real model isn't loaded yet."
+                "content": f"Placeholder response. Received: {data['messages'][-1]['content']}"
             },
             "finish_reason": "stop"
         }],
@@ -24,11 +28,15 @@ def chat_completions(message):
 
 demo = gr.Interface(
     fn=chat_completions,
-    inputs="json",
-    outputs="json",
+    inputs=None,
+    outputs=None,
     title="Chat Completions API",
     description="Send a POST request to /v1/chat/completions"
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.launch()
+
+demo.queue()
+demo.launch()
+api = gr.mount_gradio_app(demo, "/v1/chat/completions", api_name="chat_completions")