Using Trainer with do_train + do_eval

Hi, I’m currently training a T5 model using the HF Trainer and looking through the TrainingArguments documentation as well as example scripts.

At the moment I am trying to understand the use of the the two optional arguments within TrainingArguments ‘do_train’ and ‘do_eval’. Some examples notebooks don’t specify these, and therefore default these to false.

From experimenting, you can run trainer.evaluate() without ‘do_eval’, and I can’t figure out what ‘do_train’ would do if it defaults to False


Those arguments are only used by the scripts, not the Trainer itself.