For fine-tuned LLAMA 2

minimum how many records of the dataset (row) in CSV for fine-tuned LLAMA 2?

on average what size of dataset to fine tunned llama 2 in best case