Hugging Face's logo
language: ig datasets:
bert-base-multilingual-cased-finetuned-igbo
Model description
bert-base-multilingual-cased-finetuned-igbo is a Igbo BERT model obtained by fine-tuning bert-base-multilingual-cased model on Igbo 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 Igbo corpus.
Intended uses & limitations
How to use
You can use this model with Transformers pipeline for masked token prediction.
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='Davlan/bert-base-multilingual-cased-finetuned-igbo')
>>> unmasker("Reno Omokri na Gọọmentị [MASK] enweghị ihe ha ga-eji hiwe ya bụ mmachi.")
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 JW300 + OPUS CC-Align + IGBO NLP Corpus +Igbo CC-100
Training procedure
This model was trained on a single NVIDIA V100 GPU
Eval results on Test set (F-score, average over 5 runs)
Dataset | mBERT F1 | ig_bert F1 |
---|---|---|
MasakhaNER | 85.11 | 86.75 |
BibTeX entry and citation info
By David Adelani
- Downloads last month
- 7