File size: 2,845 Bytes
9103c00 |
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 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 |
Hugging Face's logo
---
language: yo
datasets:
---
# bert-base-multilingual-cased-finetuned-yoruba
## Model description
**bert-base-multilingual-cased-finetuned-yoruba** is a **Yoruba BERT** model obtained by fine-tuning **bert-base-multilingual-cased** model on Yorùbá language texts. It provides **better performance** than the multilingual BERT on text classification and named entity recognition datasets.
Specifically, this model is a *bert-base-multilingual-cased* model that was fine-tuned on Yorùbá corpus.
## Intended uses & limitations
#### How to use
You can use this model with Transformers *pipeline* for masked token prediction.
```python
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='Davlan/bert-base-multilingual-cased-finetuned-yoruba')
>>> unmasker("Arẹmọ Phillip to jẹ ọkọ [MASK] Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun")
[{'sequence': '[CLS] Arẹmọ Phillip to jẹ ọkọ Mary Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun [SEP]', 'score': 0.1738305538892746,
'token': 12176,
'token_str': 'Mary'},
{'sequence': '[CLS] Arẹmọ Phillip to jẹ ọkọ Queen Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun [SEP]', 'score': 0.16382873058319092,
'token': 13704,
'token_str': 'Queen'},
{'sequence': '[CLS] Arẹmọ Phillip to jẹ ọkọ ti Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun [SEP]', 'score': 0.13272495567798615,
'token': 14382,
'token_str': 'ti'},
{'sequence': '[CLS] Arẹmọ Phillip to jẹ ọkọ King Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun [SEP]', 'score': 0.12823280692100525,
'token': 11515,
'token_str': 'King'},
{'sequence': '[CLS] Arẹmọ Phillip to jẹ ọkọ Lady Elizabeth to ti wa lori aisan ti dagbere faye lẹni ọdun mọkandilọgọrun [SEP]', 'score': 0.07841219753026962,
'token': 14005,
'token_str': 'Lady'}]
```
#### Limitations and bias
This model is limited by its training dataset of entity-annotated news articles from a specific span of time. This may not generalize well for all use cases in different domains.
## Training data
This model was fine-tuned on Bible, JW300, [Menyo-20k](https://huggingface.co/datasets/menyo20k_mt), [Yoruba Embedding corpus](https://huggingface.co/datasets/yoruba_text_c3) and [CC-Aligned](https://opus.nlpl.eu/), Wikipedia, news corpora (BBC Yoruba, VON Yoruba, Asejere, Alaroye), and other small datasets curated from friends.
## Training procedure
This model was trained on a single NVIDIA V100 GPU
## Eval results on Test set (F-score)
Dataset|F1-score
-|-
Yoruba GV NER |75.34
MasakhaNER |80.82
BBC Yoruba |80.66
### BibTeX entry and citation info
By David Adelani
```
```
|