Update app.py

app.py

@@ -6,7 +6,6 @@ import spaces
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
 from typing import List, Dict, Optional, Tuple
-from http import HTTPStatus
 
 DESCRIPTION = """
 # QwQ Distill
@@ -88,7 +87,7 @@ def generate(
     top_p: float = 0.9,
     top_k: int = 50,
     repetition_penalty: float = 1.2,
-) -> Iterator[Tuple[str, List, str]]:
+) -> Iterator[str]:
     if query is None:
         query = ''
     if history is None:
@@ -135,11 +134,7 @@ def generate(
     outputs = []
     for text in streamer:
         outputs.append(text)
-        response = "".join(outputs)
-        # Update history with the new response
-        new_messages = messages + [{'role': Role.ASSISTANT, 'content': response}]
-        system, new_history = messages_to_history(new_messages)
-        yield "", new_history, system
+        yield "".join(outputs)
 
 
 demo = gr.ChatInterface(