The repo in based on LLaMA2-Accessory. Follow the instructions to install. Then replace the accessory folder with the one in this repo. Model checkpoint is at Huggingface.

Run the following command in data_utils:

python amex_to_qa.py --level all --root-dir /path/to/AMEX

You can change the level to 'l1', 'l2', 'l3', 'all' to process different levels of data.

Follow the instructions to finetune the model.

Follow the instructions to inference. Or run the following command in accessory:

bash infer.sh