Support for LoRA training from raw text file?

[araleza]

Hi,

I've previously been using oobabooga's Training tab to make LoRAs, and I wanted to try using LLaMA-Factory instead. I have a text file with a lot of speech from a single person that I want to train a model to sound like.

But, I can't see how to get LLaMa-Factory to read this file. It seems to only want conversational exchanges between two people, but there is only one person speaking in my text file.

Is this possible with LLaMA-Factory? If not, could I request that as a feature?

Thank you.

[gotzmann]

From what I know, you should place your dataset file somewhere on the filesystem, then just add name and path to your dataset into data/dataset_info.json file (similar to wikidemo there) and then train LoRA as specified in README.