Fine-tuning an LLM on non-Q&A and unlabeled dataset

Short Course Q&A Finetuning Large Language Models
1

In the course, it was mentioned that we can fine-tune an LLM on a non-Q&A dataset. To do this, we need to convert the data to a Q&A dataset using different techniques, such as using another model.

How can we fine-tune an LLM on a dataset of documents (which are not in Q&A format and are unlabeled)?