Finetune LLaMA2 model with datasets missing labels

Hello,

I’ve reviewed examples of fine-tuning the LLaMA2 model, but it appears that only labeled datasets are used for fine-tuning.

I’m curious if I can fine-tune the LLaMA2 model to learn the style of text (e.g., all the texts in the datasets are from Shakespeare).

In particular, I believe using autotrain-advanced is an effective tool for fine-tuning, but it necessitates a specific text format (i.e., ### Human: xxx ### Assistant: xxx).

Any comments are welcome.