Training Arguments - eval_step vs save_step

Unless you are using another argument, the save doesn’t care about the best model, so will just save every save_steps regardless of which step had the better loss.

2 Likes