When training, the default value of the argument of log_dir is None, resulting no checkpoint saved every epoch, which is inconvenient for users. I'd like to recommend to set the default value of log_dir, such as mxfold_log, and create it (os.makedirs("mxfold_log", exist_ok=True)).

The origin code for saving checkpoints every epoch is as follows:

if args.log_dir is not None:
    self.save_checkpoint(args.log_dir, epoch)