Update app.py

app.py

@@ -62,7 +62,7 @@ def calculate_eou(chat_ctx, session):
 def respond(
     message,
     history: list[tuple[str, str]],
-    max_tokens=1024,
+	max_tokens=2048,
     temperature=0.6,
     top_p=0.95,
 ):
@@ -89,12 +89,12 @@ def respond(
         stream=True,
         temperature=temperature,
         top_p=top_p,
+        stream=True
     )
 
     for chunk in stream:
-        token = chunk.choices[0].delta.content or ""
-        full_response += token
-        yield full_response  # Send raw unmodified response to Gradio
+        print(chunk.choices[0].delta.content)
+        yield chunk.choices[0].delta.content  # Send raw unmodified response to Gradio
 
     # This will match both console and Gradio output
     print(f"Final response: {full_response}")