Does the model learn from the eval_dataset when fine-tuning?

I have been fine-tuning the BART (BartForConditionalGeneration) model for summarization in a specific case, and I was wondering if the model also gets trained on the eval_dataset, or only on the train_dataset. In other words, if I test the model afterwards, can I still use the eval_dataset for testing and measuring how well the model performs, or has it ‘seen’ it in its training already?

Only the train_dataset is used for training, so unless you have some overlap between the train_dataset and the eval_dataset you passed to the Trainer, the answer is no.

1 Like

Great, thank you so much for your answer!