---
license: mit
language:
- fas
tags:
- tokenizer
- unigram
- flexitok
- fineweb2
---

# UnigramLM Tokenizer: fas_Arab (32K)

A **UnigramLM** tokenizer trained on **fas_Arab** data from Fineweb-2-HQ.

## Training Details

| Parameter | Value |
|-----------|-------|
| Algorithm | UnigramLM |
| Language | `fas_Arab` |
| Target Vocab Size | 32,000 |
| Final Vocab Size | 0 |
| Pre-tokenizer | ByteLevel |
| Normalizer | NFC |
| Special Tokens | `<s>`, `</s>`, `<pad>`, `<unk>` |
| Training Shards | 2 |

## Usage

```python
from transformers import AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("<repo_id>")
tokens = tokenizer.encode("Hello, world!")
```

## Files

- `tokenizer.json` — Full HuggingFace tokenizer
- `vocab.json` — Vocabulary mapping
- `tokenizer.model` — SentencePiece protobuf (if available)