metadata
license: mit
language:
- en
tags:
- gender bias
- debias
- fine-tune
- gpt2
- llm
Top-10 attention heads causing gender bias have been identified using DiffMask+, as proposed in the paper "Identifying and Adapting Transformer-Components Responsible for Gender Bias in an English Language Model." The model was fine-tuned using the Balanced BUG dataset (Levy et al., 2021).