### Code for Debiasing as Nonlinear Integer Programming ###

# Prerequisites
Please obtain Huggingface Llama-2 models for evaluations on Llama-2-7B and Llama-2-13B (https://huggingface.co/meta-llama).
Install the required packages via pip:

pip install -r requirements.txt


# Run experiments
The per-class output probabilities are located under ./vectors

Perform DNIP optimization and evaluation, go to project directory and run:

python dnip.py \
	-c config/default_params.json \
	--vec_dir ${vector_dir}

E.g.,
python dnip.py \
	-c config/default_params.json \
	--vec_dir vectors/vecs_gpt2-xl/dbpedia_gpt2-xl_shot1_seed0


