|
--- |
|
license: cc-by-nc-sa-4.0 |
|
pipeline_tag: fill-mask |
|
language: en |
|
datasets: |
|
- OpenSubtitles |
|
library_name: transformers |
|
--- |
|
## Model description |
|
This model is based on [An Exploration of Hierarchical Attention Transformers for Efficient Long Document Classification](https://arxiv.org/abs/2210.05529). Ilias Chalkidis, Xiang Dai, Manos Fergadiotis, Prodromos Malakasiotis, and Desmond Elliott. 2022. arXiv:2210.05529 (Preprint). |
|
|
|
Initial weights were taken from [google/bert_uncased_L-8_H-256_A-4](https://huggingface.co/google/bert_uncased_L-8_H-256_A-4). |
|
Model was additionally pretrained for 20_000 steps on 5m lines of text from english version of [OpenSubtitles](http://www.opensubtitles.org/) dataset. |
|
|
|
Maximum input length is 512 tokens that is enoungh to encode dialog with few previous utterances (average sentence length per utterance in SWDA, MAPTASK, MRDA, BT_OASIS, FRAMES, AMI, DSTC3 is less than 11 tokens). |