The following models have been made available by the BLURB team.
This is the primary model proposed and evaluated in our publication.
To generate this model, we extended the pretraining of a BERT Base model over a collection of PubMed abstracts, which improves the model’s ability to generalize to a wide range of biomedical natural language tasks.
🤗 BiomedBERT (abstracts only)
This model was proposed and evaluated in Section 6.4 of our publication. It is an uncased BERT Base model that was pretrained over PubMed abstracts and full PubMed Central articles. At the time of release, this model achieves state-of-the-art performance for our BLURB Score.
🤗 BiomedBERT (abstracts + full text)