This repisotory has the files for a possible publication about social bias quantification in Large Language Models (LLMs). The models used are: BERT, GPT-2, LLaMA-2-7B, Falcon-7B, and MistralAI-7B. The fine-tuned models with the highest classification accuracy are on my page on huggingface https://huggingface.co/kumo24. The results for LLaMA-2 is excluded because of the restrictions on its use.

The code works on Python 3.9 and later versions, check requirements.txt for the required packages.

python run.py