Hey,
I am finetuning a T5 model right now and would like to adjust the Teacher Forcing.
Rather than using the ground truth as an input for the next token I would prefer to use more of the models predictions as an input.
Is this possible with huggingface? Thank you.