Specifically, when I try to specify save_strategy=epoch in TrainingArguments, I get the error message in the issue title. I tried updating to a more recent version of Transformers as per another issue, but that didn't work. I'm not sure what to do about it. Collaborator sgugger ...
I'm usingBERTForMaskedLM, but it doesn't matter, because it's a generalTrainerissue. Perhapssave_strategyandevaluation_strategyshould both have the same default value? To reproduce Steps to reproduce the behavior: InTrainer, setload_best_model_at_endtoTruewithout assigning any values tosave_str...