Stereotype and Skew: Quantifying Gender Bias in Pre-trained and Fine-tuned Language Models