Mathematical Structure Aware BERT

Pretrained model based on bert-base-cased with further mathematical pre-training.

Compared to bert-base-cased, 300 additional mathematical LaTeX tokens have been added before the mathematical pre-training.