Update app.py

app.py

@@ -102,7 +102,7 @@ def bot_streaming(message, history):
     streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
     
     image_tensor = model.process_images([image], model.config).to(dtype=model.dtype)
-    generation_kwargs = dict(input_ids=input_ids, images=image_tensor, streamer=streamer, max_new_tokens=100, stopping_criteria=[stopping_criteria])
+    generation_kwargs = dict(input_ids=input_ids.to('cuda'), images=image_tensor, streamer=streamer, max_new_tokens=100, stopping_criteria=[stopping_criteria])
     generated_text = ""
     thread = Thread(target=model.generate, kwargs=generation_kwargs)
     thread.start()