- numpy
- torch
- transformers
We release the debiased model here.
python generate_prompts.py \
--debias_type gender or race
--model_type bert or roberta or albert
--model_name_or_path bert-base-uncased, etc
python auto-debias.py
--debias_type gender or race
--model_type bert or roberta or albert
--model_name_or_path bert-base-uncased, etc
--prompts_file prompts_bert-base-uncased_gender
We run the SEAT using the code from Liang et al.
We run GLUE from transformers.