How Hugging Face is tackling bias in NLP