Key Error "loss"

Hi everyone,
I am having issue when I try to fine tune a pre-trained model. In particular I am trying to use OPT model by Meta to perform NMT. First of all I created the tokenizer and following the instructions of the documentation I used the AutoTokenizer (with flag, use_fast=False) or GPT2Tokenizer. Then I use Seq2SeqTrainer and when I try to train the model I get the error Key Loss “loss”.

Reading the other topics on the forum I have seen that the problem can be that I do not have the labels of the dataset, but in reality I have them.

Furthermore I upload the notebook related to the problem that can be viewed by everyone.
Link Notebook: translation.ipynb - Google Drive