How to prevent too many checkpoints with run_clm.py?

You need to pass along --load_best_model_at_end then, so the Trainer keeps track of the best model and does not delete it.

1 Like