Update app.py

app.py

@@ -105,7 +105,7 @@ def bot_streaming(message, history):
     image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
     generation_kwargs = dict(input_ids=input_ids.to('cuda'), 
                              images=image_tensor.to('cuda'), 
-                             streamer=streamer, max_new_tokens=256, 
+                             streamer=streamer, max_new_tokens=128, 
                              stopping_criteria=[stopping_criteria], 
                              temperature=0.25,
                              top_k=40,