Update app.py

app.py

@@ -66,7 +66,8 @@ def stream_chat(message, history: list, temperature: float, max_new_tokens: int)
     
     inputs = processor.tokenizer.apply_chat_template(conversation, tokenize=False, add_generation_prompt=True)
     inputs_ids = processor(inputs, image, return_tensors="pt").to(0)
-    
+    streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": True, "skip_prompt": True, 'clean_up_tokenization_spaces':False,}) 
+
     generate_kwargs = dict(
         streamer=streamer,
         max_new_tokens=max_new_tokens,
@@ -78,7 +79,6 @@ def stream_chat(message, history: list, temperature: float, max_new_tokens: int)
         generate_kwargs["do_sample"] = False
     generate_kwargs = {**inputs_ids, **generate_kwargs}
     
-    streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": True, "skip_prompt": True, 'clean_up_tokenization_spaces':False,}) 
 
     thread = Thread(target=model.generate, kwargs=generate_kwargs)
     thread.start()