AIMH/mental-longformer-base-4096

mental-longformer-base-4096 is a model pretrained from the checkpoint of longformer-base-4096 for the mental healthcare domain. Longformer is a transformer model for long documents. longformer-base-4096 is a BERT-like model started from the RoBERTa checkpoint and pretrained for MLM on long documents. It supports sequences of length up to 4,096.

Usage

Load the model via Huggingface’s Transformers library:

from transformers import LongformerTokenizer, LongformerModel
tokenizer = LongformerTokenizer.from_pretrained("AIMH/mental-longformer-base-4096")
model = LongformerModel.from_pretrained("AIMH/mental-longformer-base-4096")

To minimize the influence of worrying mask predictions, this model is gated. To download a gated model, you’ll need to be authenticated. Know more about gated models.

Paper

@article{ji-domain-specific,
  author        = {Shaoxiong Ji and Tianlin Zhang and Kailai Yang and Sophia Ananiadou and Erik Cambria and J{\"o}rg Tiedemann},
  journal       = {arXiv preprint arXiv:2304.10447},
  title         = {Domain-specific Continued Pretraining of Language Models for Capturing Long Context in Mental Health},
  year          = {2023},
  url           = {https://arxiv.org/abs/2304.10447}
}

Disclaimer

The model predictions are not psychiatric diagnoses. We recommend anyone who suffers from mental health issues to call the local mental health helpline and seek professional help if possible.