Update app.py

app.py

@@ -170,7 +170,7 @@ with gr.Blocks() as demo:
                         inputs=[image_input, query_input, assistant_prefix, decoding_strategy, temperature,
                                                               max_new_tokens, repetition_penalty, top_p],
                         outputs=output,
-                        fn=model_inference
+                        fn=model_inference, cache_examples=False
                     )
         
 
@@ -178,4 +178,4 @@ with gr.Blocks() as demo:
                                                       max_new_tokens, repetition_penalty, top_p], outputs=output)
 
 
-demo.launch(debug=True, cache_examples=False)
+demo.launch(debug=True)