File size: 1,465 Bytes
ed3ae24 1869669 ed3ae24 1869669 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 |
---
title: README
emoji: 🏃
colorFrom: gray
colorTo: purple
sdk: static
pinned: false
---
# Model Description
ClinicalMobileBERT is the result of training the [BioMobileBERT](https://huggingface.co/google/nlpie/bio-mobilebert) model in a continual learning scenario for 3 epochs using a total batch size of 192 on the MIMIC-III notes dataset.
# Initialisation
We initialise our model with the pre-trained checkpoints of the [BioMobileBERT](https://huggingface.co/google/nlpie/bio-mobilebert) model available on Huggingface.
# Architecture
MobileBERT uses a 128-dimensional embedding layer followed by 1D convolutions to up-project its output to the desired hidden dimension expected by the transformer blocks. For each of these blocks, MobileBERT uses linear down-projection at the beginning of the transformer block and up-projection at its end, followed by a residual connection originating from the input of the block before down-projection. Because of these linear projections, MobileBERT can reduce the hidden size and hence the computational cost of multi-head attention and feed-forward blocks. This model additionally incorporates up to four feed-forward blocks in order to enhance its representation learning capabilities. Thanks to the strategically placed linear projections, a 24-layer MobileBERT (which is used in this work) has around 25M parameters.
# Citation
If you use this model, please consider citing the following paper:
```bibtex
``` |