We have two independent codebases, one for the analysis, which is most of the paper, and one for traning and finetuning. Each of the subfolders has its own readme file. Please check those for more details.