File size: 1,585 Bytes
52978a1
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
---
language: en
tags: deberta
thumbnail: https://huggingface.co/front/thumbnails/microsoft.png
license: mit
---

## DeBERTa: Decoding-enhanced BERT with Disentangled Attention

[DeBERTa](https://arxiv.org/abs/2006.03654) improves the BERT and RoBERTa models using disentangled attention and enhanced mask decoder. With those two improvements, DeBERTa out perform RoBERTa on a majority of NLU tasks with 80GB training data. 

Please check the [official repository](https://github.com/microsoft/DeBERTa) for more details and updates.

This is the DeBERTa V3 small model with 6 layers, 768 hidden size. Total parameters is 143M while Embedding layer take about 98M due to the usage of 128k vocabulary. It's trained with 160GB data.
For more details of our V3 model, please check appendix A11 in our paper.

#### Fine-tuning on NLU tasks

We present the dev results on SQuAD 1.1/2.0 and MNLI tasks.
| Model             | SQuAD 1.1 | SQuAD 2.0 | MNLI-m |
|-------------------|-----------|-----------|--------|
| RoBERTa-base      | 91.5/84.6 | 83.7/80.5 | 87.6   |
| XLNet-base        | -/-       | -/80.2    | 86.8   |
| **DeBERTa-v3-small**  | 93.1/87.2 | 86.2/83.1 | 88.2   |


### Citation

If you find DeBERTa useful for your work, please cite the following paper:

``` latex
@inproceedings{
he2021deberta,
title={DEBERTA: DECODING-ENHANCED BERT WITH DISENTANGLED ATTENTION},
author={Pengcheng He and Xiaodong Liu and Jianfeng Gao and Weizhu Chen},
booktitle={International Conference on Learning Representations},
year={2021},
url={https://openreview.net/forum?id=XPZIaotutsD}
}
```