Create README.md
Browse files
README.md
ADDED
@@ -0,0 +1,53 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# gaBERT
|
2 |
+
|
3 |
+
[gaBERT](https://arxiv.org/abs/2107.12930) is a BERT-base model trained on 7.9M Irish sentences. For more details, including the hyperparameters and pretraining corpora used please refer to our paper.
|
4 |
+
|
5 |
+
### How to use gaBERT with HuggingFace
|
6 |
+
|
7 |
+
```
|
8 |
+
from transformers import AutoModelWithLMHead, AutoTokenizer
|
9 |
+
import torch
|
10 |
+
|
11 |
+
tokenizer = AutoTokenizer.from_pretrained("DCU-NLP/bert-base-irish-cased-v1")
|
12 |
+
model = AutoModelWithLMHead.from_pretrained("DCU-NLP/bert-base-irish-cased-v1")
|
13 |
+
|
14 |
+
sequence = f"Ceoltóir {tokenizer.mask_token} ab ea Johnny Cash."
|
15 |
+
|
16 |
+
input = tokenizer.encode(sequence, return_tensors="pt")
|
17 |
+
mask_token_index = torch.where(input == tokenizer.mask_token_id)[1]
|
18 |
+
|
19 |
+
token_logits = model(input)[0]
|
20 |
+
mask_token_logits = token_logits[0, mask_token_index, :]
|
21 |
+
|
22 |
+
top_5_tokens = torch.topk(mask_token_logits, 5, dim=1).indices[0].tolist()
|
23 |
+
|
24 |
+
for token in top_5_tokens:
|
25 |
+
print(sequence.replace(tokenizer.mask_token, tokenizer.decode([token])))
|
26 |
+
```
|
27 |
+
|
28 |
+
### BibTeX entry and citation info
|
29 |
+
|
30 |
+
If you use this model in your research, please consider citing our paper:
|
31 |
+
|
32 |
+
```
|
33 |
+
@article{DBLP:journals/corr/abs-2107-12930,
|
34 |
+
author = {James Barry and
|
35 |
+
Joachim Wagner and
|
36 |
+
Lauren Cassidy and
|
37 |
+
Alan Cowap and
|
38 |
+
Teresa Lynn and
|
39 |
+
Abigail Walsh and
|
40 |
+
M{\'{\i}}che{\'{a}}l J. {\'{O}} Meachair and
|
41 |
+
Jennifer Foster},
|
42 |
+
title = {gaBERT - an Irish Language Model},
|
43 |
+
journal = {CoRR},
|
44 |
+
volume = {abs/2107.12930},
|
45 |
+
year = {2021},
|
46 |
+
url = {https://arxiv.org/abs/2107.12930},
|
47 |
+
archivePrefix = {arXiv},
|
48 |
+
eprint = {2107.12930},
|
49 |
+
timestamp = {Fri, 30 Jul 2021 13:03:06 +0200},
|
50 |
+
biburl = {https://dblp.org/rec/journals/corr/abs-2107-12930.bib},
|
51 |
+
bibsource = {dblp computer science bibliography, https://dblp.org}
|
52 |
+
}
|
53 |
+
```
|