metadata
license: apache-2.0
language:
- ti
- am
- ar
| Model | Vocabulary Size |
|---|---|
| XLM-Roberta | 250002 |
| EXLMR | 280147 |
Model Card
The EXLMR model is a multilingual transformer that expands the XLM-RoBERTa tokenizer by adding vocabulary for low-resource languages such as Tigrinya and Amharic. It solves issues like out-of-vocabulary words and over-tokenization, enhancing the model's ability to represent languages written in the Ge'ez script. The model can be fine-tuned for various multilingual tasks, including sentiment analysis, question answering, named entity recognition, and paraphrase detection. These improvements make EXLMR highly effective for low-resource languages, while still supporting a broad range of languages with strong overall performance.