Update app.py

app.py

@@ -32,7 +32,7 @@ async def respond(
         messages.append({"role": "user", "content": message})
 
         response = ""
-        # Use client to stream chat completions properly
+        # Properly stream chat completions using dot notation
         stream = client.chat.completions.create(
             model="NousResearch/Hermes-3-Llama-3.1-8B",
             max_tokens=max_tokens,
@@ -41,9 +41,10 @@ async def respond(
             top_p=top_p,
             messages=messages,
         )
-        for message in stream:  # Iterate over the streamed response
-            token = message['choices'][0]['delta'].get('content', '')
-            response += token
+        for chunk in stream:  # Iterate over the streamed response chunks
+            if hasattr(chunk.choices[0].delta, 'content'):
+                token = chunk.choices[0].delta.content
+                response += token
         return response
     except APIError as e:
         error_details = e.body