|
--- |
|
license: cc-by-sa-4.0 |
|
base_model: indobenchmark/indobert-large-p2 |
|
language: |
|
- min |
|
- ban |
|
- bug |
|
- id |
|
pretty_name: IndoBERTNusa |
|
tags: |
|
- generated_from_trainer |
|
datasets: |
|
- prosa-text/nusa-dialogue |
|
- indonlp/NusaX-MT |
|
pipeline_tag: fill-mask |
|
--- |
|
|
|
# IndoBERTNusa (IndoBERT Adapted for Balinese, Buginese, and Minangkabau) |
|
|
|
This repository contains a language adaptation and fine-tuning of the Indobenchmark IndoBERT language model for three specific languages: Balinese, Buginese, and Minangkabau. |
|
The adaptation was performed using [nusa-translation](https://huggingface.co/datasets/prosa-text/nusa-translation) dataset. |
|
|
|
## Model Details |
|
|
|
- **Base Model**: [indobenchmark/indobert-large-p2](https://huggingface.co/indobenchmark/indobert-large-p2) |
|
- **Adaptation Data**: [nusa-translation](https://huggingface.co/datasets/prosa-text/nusa-translation) |
|
|
|
|
|
## Performance Comparison / Benchmark |
|
|
|
### Topic Classification |
|
|
|
We tested the model after it was fine-tuned for topic classification using [nusa-dialogue](https://huggingface.co/datasets/prosa-text/nusa-dialogue) dataset. |
|
|
|
|
|
| Language | indobert-large-p2 (F1) | indobert-nusa (F1) | |
|
|-------------|------------------------|------------------------| |
|
| Balinese | 82.37 | **84.23** | |
|
| Buginese | 80.53 | **82.03** | |
|
| Minangkabau | 84.49 | **86.30** | |
|
|
|
|
|
### Language Identification |
|
|
|
We also tested the model after it was fine-tuned for language identification using [nusaX](https://github.com/IndoNLP/nusax) dataset. |
|
|
|
| Model | F1-score | |
|
|----------------------|--------------| |
|
| indobert-large-p2 | 98.21 | |
|
| **indober-nusa** | **98.45** | |
|
|
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 5e-05 |
|
- train_batch_size: 8 |
|
- eval_batch_size: 8 |
|
- seed: 42 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 3.0 |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.33.1 |
|
- Pytorch 2.1.2+cu121 |
|
- Datasets 2.16.1 |
|
- Tokenizers 0.13.3 |
|
|
|
|
|
## Additional Information |
|
|
|
### Licensing Information |
|
The dataset is released under the terms of **CC-BY-SA 4.0**. |
|
By using this model, you are also bound to the respective Terms of Use and License of the dataset. |
|
For commercial use in small businesses and startups, please contact us (business@prosa.ai) for permission to use the datasets by informing company profile and propose of usage. |
|
|
|
### Acknowledgement |
|
This research work is funded and supported by The Deutsche Gesellschaft für Internationale Zusammenarbeit (GIZ) GmbH and FAIR Forward - Artificial Intelligence for all. We thank Direktorat Jenderal Pendidikan Tinggi, Riset, dan Teknologi Kementerian Pendidikan, Kebudayaan, Riset, dan Teknologi (Ditjen DIKTI) for providing the computing resources for this project. |
|
|
|
### Contact Us |
|
If you have any question please contact our support team at `business@prosa.ai`. |
|
|
