ScholarBERT_10_WB Model

This is the ScholarBERT_10_WB variant of the ScholarBERT model family.

The model is pretrained on a large collection of scientific research articles (22.1B tokens). Additionally, the pretraining data also includes the Wikipedia+BookCorpus, which are used to pretrain the BERT-base and BERT-large models.

This is a cased (case-sensitive) model. The tokenizer will not convert all inputs to lower-case by default.

The model is based on the same architecture as BERT-large and has a total of 340M parameters.

Model Architecture

Hyperparameter	Value
Layers	24
Hidden Size	1024
Attention Heads	16
Total Parameters	340M

Training Dataset

The vocab and the model are pertrained on 10% of the PRD scientific literature dataset and Wikipedia+BookCorpus.

The PRD dataset is provided by Public.Resource.Org, Inc. (“Public Resource”), a nonprofit organization based in California. This dataset was constructed from a corpus of journal article files, from which We successfully extracted text from 75,496,055 articles from 178,928 journals. The articles span across Arts & Humanities, Life Sciences & Biomedicine, Physical Sciences, Social Sciences, and Technology. The distribution of articles is shown below.

BibTeX entry and citation info

If using this model, please cite this paper:

@misc{hong2023diminishing,
      title={The Diminishing Returns of Masked Language Models to Science}, 
      author={Zhi Hong and Aswathy Ajith and Gregory Pauloski and Eamon Duede and Kyle Chard and Ian Foster},
      year={2023},
      eprint={2205.11342},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}