Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,56 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Fulfulde Tokenizer for NLP tasks
|
| 2 |
+
|
| 3 |
+
## Model Description
|
| 4 |
+
This tokenizer was developed for Fulfulde, a language from the fula[ni] family of languages in Cameroon. The tokenizer is based on the WordPiece model architecture and has been fine-tuned to handle the unique phonetic and diacritical features of the Fulfulde language.
|
| 5 |
+
|
| 6 |
+
- **Developed by**: DS4H-ICTU Research Group in Cooperation with the
|
| 7 |
+
- **Language(s)**: Fulfulde Adamaoua (Fula[ni] language from Cameroon)
|
| 8 |
+
- **License**: Apache 2.0 (or specify if different)
|
| 9 |
+
- **Model Type**: Tokenizer (WordPiece)
|
| 10 |
+
|
| 11 |
+
## Model Sources
|
| 12 |
+
- **Repository**: [Your repository URL]
|
| 13 |
+
- **Paper**: [Link to related paper if available]
|
| 14 |
+
- **Demo**: [Optional: link to demo]
|
| 15 |
+
|
| 16 |
+
## Uses
|
| 17 |
+
- **Direct Use**: This tokenizer is designed for NLP tasks such as Named Entity Recognition (NER), translation, and text generation in the Fulfulde Adamaoua language.
|
| 18 |
+
- **Downstream Use**: Can be used as a foundation for models processing Fulfulde Adamaoua text.
|
| 19 |
+
|
| 20 |
+
## Bias, Risks, and Limitations
|
| 21 |
+
- **Biases**: The tokenizer might not perfectly capture linguistic nuances due to the limited size of the Fulfulde corpus.
|
| 22 |
+
- **Out-of-Scope Use**: The tokenizer may not perform well for non-Fulfulde languages.
|
| 23 |
+
|
| 24 |
+
## Training Details
|
| 25 |
+
- **Training Data**: Extracted from YambeFulfulde Adamaoua Bible text corpus (final_dataset.xlsx).
|
| 26 |
+
- **Training Procedure**: Preprocessing of text involved normalization of diacritics, tokenization using WordPiece, and post-processing to handle special tokens.
|
| 27 |
+
- **Training Hyperparameters**:
|
| 28 |
+
- Vocabulary Size: 25,000
|
| 29 |
+
- Special Tokens: [UNK], [PAD], [CLS], [SEP], [MASK]
|
| 30 |
+
|
| 31 |
+
## Evaluation
|
| 32 |
+
- **OOV Rate**: 0.00%
|
| 33 |
+
- **Tokenization Efficiency**: Average tokens per sentence: 26.884630350194552
|
| 34 |
+
- **Special Character Handling**: Successfully handles diacritics and tone markers in Fulfulde Adamaoua.
|
| 35 |
+
|
| 36 |
+
## Environmental Impact
|
| 37 |
+
- **Hardware Type**: Google Colab GPU
|
| 38 |
+
- **Hours Used**: 4 hours (training time)
|
| 39 |
+
- **Cloud Provider**: Google Cloud
|
| 40 |
+
- **Carbon Emitted**: Estimated using [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700) calculator
|
| 41 |
+
|
| 42 |
+
## Citation
|
| 43 |
+
If you use this tokenizer in your work, please cite it using the following format:
|
| 44 |
+
|
| 45 |
+
```
|
| 46 |
+
@misc{fulfulde_tokenizer,
|
| 47 |
+
title = {Fulfulde Adamaoua Tokenizer},
|
| 48 |
+
author = {Ing. Zingui Fred Mike},
|
| 49 |
+
year = {2024},
|
| 50 |
+
publisher = {Hugging Face},
|
| 51 |
+
url = {https://huggingface.co/FredMike23/tokenizer-Fulfulde_adamaoua}
|
| 52 |
+
}
|
| 53 |
+
```
|
| 54 |
+
|
| 55 |
+
## Contact Information
|
| 56 |
+
For more information, contact the developers at: mikezingui@yahoo.com
|