Commit
•
df8bdb4
1
Parent(s):
da7fbc0
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language: wo
|
| 3 |
+
tags:
|
| 4 |
+
- bert
|
| 5 |
+
- language-model
|
| 6 |
+
- wo
|
| 7 |
+
- wolof
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
+
# Soraberta: Unsupervised Language Model Pre-training for Wolof
|
| 11 |
+
|
| 12 |
+
**bert-base-wolof** is pretrained bert-base model on wolof language .
|
| 13 |
+
|
| 14 |
+
## Soraberta models
|
| 15 |
+
|
| 16 |
+
| Model name | Number of layers | Attention Heads | Embedding Dimension | Total Parameters |
|
| 17 |
+
| :------: | :---: | :---: | :---: | :---: |
|
| 18 |
+
| `bert-base` | 6 | 12 | 514 | 56931622 M |
|
| 19 |
+
|
| 20 |
+
|
| 21 |
+
|
| 22 |
+
|
| 23 |
+
## Using Soraberta with Hugging Face's Transformers
|
| 24 |
+
|
| 25 |
+
|
| 26 |
+
```python
|
| 27 |
+
>>> from transformers import pipeline
|
| 28 |
+
>>> unmasker = pipeline('fill-mask', model='abdouaziiz/soraberta')
|
| 29 |
+
>>> unmasker("juroom naari jullit man nanoo boole jend aw nag walla <mask>.")
|
| 30 |
+
|
| 31 |
+
[{'sequence': 'juroom naari jullit man nanoo boole jend aw nag walla gileem.',
|
| 32 |
+
'score': 0.9783930778503418,
|
| 33 |
+
'token': 4621,
|
| 34 |
+
'token_str': ' gileem'},
|
| 35 |
+
{'sequence': 'juroom naari jullit man nanoo boole jend aw nag walla jend.',
|
| 36 |
+
'score': 0.009271537885069847,
|
| 37 |
+
'token': 2155,
|
| 38 |
+
'token_str': ' jend'},
|
| 39 |
+
{'sequence': 'juroom naari jullit man nanoo boole jend aw nag walla aw.',
|
| 40 |
+
'score': 0.0027585660573095083,
|
| 41 |
+
'token': 704,
|
| 42 |
+
'token_str': ' aw'},
|
| 43 |
+
{'sequence': 'juroom naari jullit man nanoo boole jend aw nag walla pel.',
|
| 44 |
+
'score': 0.001120452769100666,
|
| 45 |
+
'token': 1171,
|
| 46 |
+
'token_str': ' pel'},
|
| 47 |
+
{'sequence': 'juroom naari jullit man nanoo boole jend aw nag walla juum.',
|
| 48 |
+
'score': 0.0005133090307936072,
|
| 49 |
+
'token': 5820,
|
| 50 |
+
'token_str': ' juum'}]
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
## Training data
|
| 54 |
+
The data sources are [Bible OT](http://biblewolof.com/) , [WOLOF-ONLINE](http://www.wolof-online.com/)
|
| 55 |
+
[ALFFA_PUBLIC](https://github.com/getalp/ALFFA_PUBLIC/tree/master/ASR/WOLOF)
|
| 56 |
+
|
| 57 |
+
|
| 58 |
+
|
| 59 |
+
## Contact
|
| 60 |
+
|
| 61 |
+
Please contact abdouaziz@gmail.com for any question, feedback or request.
|