checkpoint will be saved twice at the end of training when save_strategy is epoch

[AaronZLT]

Reproduction

This is caused by

transformers/src/transformers/trainer_callback.py

Line 608 in dd16acb

control.should_save = True

and

transformers/src/transformers/trainer_callback.py

Line 623 in dd16acb

control.should_save = True

Saving will happen at the end of training and the end of epoch (which is also triggered when setting save_strategy to epoch). Same model will be saved twice, also, the on_save function will be triggered twice.

[NosimusAI]

Fixed by Nosimus AI - #36219