README.md

---
license: apache-2.0
language:
- ar
tags:
- SP
- Aranizer
- Arabic Tokenizer
---

# Aranizer | Arabic Tokenizer

**Aranizer** is an Arabic SentencePiece-based tokenizer designed for efficient and versatile tokenization. It features a vocabulary size of 32,000 tokens and is optimized for a fertility score of 1.803. The total number of tokens processed is 1,387,929, making it suitable for a wide range of NLP tasks.

## Features

- **Tokenizer Name**: Aranizer
- **Type**: SentencePiece tokenizer
- **Vocabulary Size**: 32,000
- **Total Number of Tokens**: 1,387,929
- **Fertility Score**: 1.803
- It supports Arabic Diacritization

## Aranizer Collection Achieved State of the Art Arabic Tokenizer

The Aranizer tokenizer has achieved state-of-the-art results on the [Arabic Tokenizers Leaderboard](https://huggingface.co/spaces/MohamedRashad/arabic-tokenizers-leaderboard) on Hugging Face. Below is a screenshot highlighting this achievement:

<img src="./lb.png" alt="Screenshot showing the Aranizer Tokenizer achieving state of the art" width="800">



## How to Use the Aranizer Tokenizer

The Aranizer tokenizer can be easily loaded using the `transformers` library from Hugging Face. Below is an example of how to load and use the tokenizer in your Python project:

```python
from transformers import AutoTokenizer

# Load the Aranizer tokenizer
tokenizer = AutoTokenizer.from_pretrained("riotu-lab/Aranizer-SP-32k")

# Example usage
text = "اكتب النص العربي"
tokens = tokenizer.tokenize(text)
token_ids = tokenizer.convert_tokens_to_ids(tokens)

print("Tokens:", tokens)
print("Token IDs:", token_ids)
```

```markdown
## Citation

@article{koubaa2024arabiangpt,
  title={ArabianGPT: Native Arabic GPT-based Large Language Model},
  author={Koubaa, Anis and Ammar, Adel and Ghouti, Lahouari and Necar, Omer and Sibaee, Serry},
  year={2024},
  publisher={Preprints}
}