Finetune LLaMA
This is a combination of the following three repos:
- https://github.com/tatsu-lab/stanford_alpaca
- https://github.com/zphang/minimal-llama
- https://github.com/randaller/llama-chat
It finetunes LLaMa using the Alpaca dataset with Peft.
You need to install the packages required by the above libraries.
Steps:
- Put
alpaca_data.json
to this directory. - Run
python3 alpaca.py
; this will create the trainingset and save to pickle. - Run
. train.sh
- Run
chat_alpaca.py
vschat_no_alpaca.py