How to format prompt for fine-tuning

#5
by ben-pfirsich - opened

Thank you for providing this model.

I would like to fine-tune it for a seq2seq language simplification task.

The input for the training would look something like this:
simplification-style + sep_token + original_text + sep_token + simplifiction

I am not sure how to use the given prompt template in this scenario.
Is it a bad idea to change the instruction after the system header?
Like this :"Du bist ein freundlicher und hilfreicher deutscher KI-Assistent und Experte in Sprachvereinfachung"

And where exactly in the prompt template should I use new line characters?

Sign up or log in to comment