ISBERT is a BERT-based uncased model trained on Information Systems text.

  • I used the full text of approx. 15K papers including their titles and abstracts for training
  • ISBERT has its own vocabulary that's built to best match the training corpus
  • WordPiece Bert Tokenizer was used
  • The model was initialized with BertForPreTraining
Downloads last month
Hosted inference API
Fill Mask

Mask token: [MASK]