This is the BERT-Medium model from Google: https://github.com/google-research/bert#bert. A BERT model with 2 layers, 128 hidden unit size, and 2 attention heads.