Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
---
|
2 |
+
|
3 |
+
language: ja
|
4 |
+
|
5 |
+
license: cc-by-sa-4.0
|
6 |
+
|
7 |
+
datasets:
|
8 |
+
|
9 |
+
- wikipedia
|
10 |
+
|
11 |
+
widget:
|
12 |
+
|
13 |
+
- text: 東京大学で[MASK]の研究をしています。
|
14 |
+
|
15 |
+
---
|
16 |
+
|
17 |
+
# ELECTRA small Japanese discriminator
|
18 |
+
|
19 |
+
This is a [ELECTRA](https://github.com/google-research/electra) model pretrained on texts in the Japanese language.
|
20 |
+
|
21 |
+
The codes for the pretraining are available at [retarfi/language-pretraining](https://github.com/retarfi/language-pretraining/tree/v1.0).
|
22 |
+
|
23 |
+
## Model architecture
|
24 |
+
|
25 |
+
The model architecture is the same as ELECTRA small in the [original ELECTRA paper](https://arxiv.org/abs/2003.10555); 12 layers, 256 dimensions of hidden states, and 4 attention heads.
|
26 |
+
|
27 |
+
## Training Data
|
28 |
+
|
29 |
+
The models are trained on the Japanese version of Wikipedia.
|
30 |
+
|
31 |
+
The training corpus is generated from the Japanese version of Wikipedia, using Wikipedia dump file as of June 1, 2021.
|
32 |
+
|
33 |
+
The corpus file is 2.9GB, consisting of approximately 20M sentences.
|
34 |
+
|
35 |
+
## Tokenization
|
36 |
+
|
37 |
+
The texts are first tokenized by MeCab with IPA dictionary and then split into subwords by the WordPiece algorithm.
|
38 |
+
|
39 |
+
The vocabulary size is 32768.
|
40 |
+
|
41 |
+
## Training
|
42 |
+
|
43 |
+
The models are trained with the same configuration as ELECTRA small in the [original ELECTRA paper](https://arxiv.org/abs/2003.10555); 128 tokens per instance, 128 instances per batch, and 1M training steps.
|
44 |
+
|
45 |
+
The size of the generator is 1/4 of the size of the discriminator.
|
46 |
+
|
47 |
+
## Citation
|
48 |
+
|
49 |
+
**There will be another paper for this pretrained model. Be sure to check here again when you cite.**
|
50 |
+
|
51 |
+
```
|
52 |
+
@inproceedings{bert_electra_japanese,
|
53 |
+
title = {Construction and Validation of a Pre-Trained Language Model
|
54 |
+
Using Financial Documents}
|
55 |
+
author = {Masahiro Suzuki and Hiroki Sakaji and Masanori Hirano and Kiyoshi Izumi},
|
56 |
+
month = {oct},
|
57 |
+
year = {2021},
|
58 |
+
booktitle = {"Proceedings of JSAI Special Interest Group on Financial Infomatics (SIG-FIN) 27"}
|
59 |
+
}
|
60 |
+
```
|
61 |
+
|
62 |
+
## Licenses
|
63 |
+
|
64 |
+
The pretrained models are distributed under the terms of the [Creative Commons Attribution-ShareAlike 4.0](https://creativecommons.org/licenses/by-sa/4.0/).
|
65 |
+
|
66 |
+
## Acknowledgments
|
67 |
+
|
68 |
+
This work was supported by JSPS KAKENHI Grant Number JP21K12010.
|