The LLaMA-1 model checkpoint trained on AQuA-RAT 90K training set.