license: llama3
model:llama3
fine-tune: reft[https://github.com/stanfordnlp/pyreft]
純qa資料集:50.0 epochs
