`xinlai/LISA-13B-llama2-v1-explanatory` model regression or incorrect user setting?

[Di Wu]

Hi, I am running the demo using xinlai/LISA-13B-llama2-v1-explanatory , but found out the model output is different than the github demo. Any idea what I have been missing or the new model did have some regression (e.g., loose some world knowledge?)