Update app.py

app.py

@@ -200,7 +200,15 @@ async def respond(
             return
     
     response = ""
-    async for token in completion:
+    async for token in streamChat({
+        "model": model_name,
+        "messages": messages,
+        "max_tokens": max_tokens,
+        "temperature": temperature,
+        "top_p": top_p,
+        "user": rnd(),
+        "stream": True
+    }):
         response += token['choices'][0]['delta'].get("content", token['choices'][0]['delta'].get("refusal", ""))
         yield f"{prefix}{response}"