Update app.py

app.py

@@ -90,7 +90,9 @@ def reply(message: str, history: list[str]) -> str:
     inputs = tok([preprocess(message)], return_tensors="pt")
     streamer = transformers.TextIteratorStreamer(tok)
 
-    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=512)
+    generation_kwargs = dict(
+        inputs, streamer=streamer, max_new_tokens=512, return_full_text=False
+    )
     thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()
     generated_text = ""