I would like to ask why “concat_text_input_output” this function needs to do this, connecting the question and answer together and putting them into the big model. Shouldn't we just include the question, The inputs_embeds section should only have samples ['text_input '], not samples ['text_output'], right?
I would like to ask why “concat_text_input_output” this function needs to do this, connecting the question and answer together and putting them into the big model. Shouldn't we just include the question, The inputs_embeds section should only have samples ['text_input '], not samples ['text_output'], right?