how does one finetune llama (or any other LLM) using mistral?

is the flow like this?

- take small dataset

- generate bigger dataset using mistral (how this is this done?)

- run LoRA to fine tune gemma extended dataset.