How to resume checkpoint?

[mrsempress]

Due to resource limitations, the program will be interrupted during training, and I want to continue training.
So I set save_strategy="epoch" in TrainingArguments() to save the checkpoint.
After the interruption, I will get the previous epoch, such as checkpoint-1166, and the files inside are: optimizer.pt/rng_state.pth/scheduler.pt/trainer_state.json/intervenable_model, where intervenable_model has a suffix of .bin.

But when I change trainer.train() to trainer.train(resume_from_checkpoint=True), I will get the error:

AttributeError: 'ReftModel' object has no attribute '_keys_to_ignore_on_save'

How can I achieve this?