The LLaMA-1 model checkpoint trained on MATH training set.