Clarification on creating hf_dataset_reference.py and using local datasets

[p024eb]

Hello,

I understand that by setting -add_dataset_ref, the script hf_dataset_reference.py, which describes the datasets used in the experiments on Hugging Face, will be converted into markdown format so that the LLM can recognize and understand the necessary data for the experiments.

My questions are:

How should I create hf_dataset_reference.py?

Is it possible to train using local datasets instead of Hugging Face datasets?

Thank you!