---
language:
- ar
tags:
- punctuation-restoration
- arabic
- pytorch
- bilstm
- text-processing
pipeline_tag: text-classification
widget:
- text: "هل تساءلت يوما عن معنى الحياة ما هي الأسئلة التي تشغل بالك"
  example_title: "Question Example"
- text: "الطقس جميل اليوم لا اعتقد انها ستمطر"
  example_title: "Statement Example"
---

# Arabic Punctuation Restoration Model (BiLSTM)

This is a **Bidirectional LSTM (BiLSTM)** model designed to restore punctuation marks in raw Arabic text. It takes unpunctuated Arabic text as input and inserts the appropriate punctuation marks.

## Model Details
- **Architecture:** BiLSTM (2 Layers, Hidden Dim 256)
- **Embeddings:** AraVec (Twitter-CBOW 300d)
- **Vocabulary Size:** ~50k words
- **Input:** Raw Arabic text (with or without diacritics)
- **Output:** Text with restored punctuation marks

## Supported Punctuation Marks
The model predicts the following punctuation marks:

| ID | Mark | Name |
|---|---|---|
| 0 | (None) | No Punctuation |
| 1 | **?** | Question Mark (؟) |
| 2 | **،** | Arabic Comma |
| 3 | **:** | Colon |
| 4 | **؛** | Arabic Semicolon |
| 5 | **!** | Exclamation Mark |
| 6 | **.** | Period / Full Stop |

## How to Use

Since this is a custom PyTorch model, you need to load the model structure and vocabulary. 

### Method 1: Using the Inference Script (Recommended)
Download the `inference.py` file from this repository to use the model easily.

```python
from huggingface_hub import hf_hub_download
import importlib.util

# 1. Download the script
script_path = hf_hub_download(repo_id="malkhuzanie/arabic-punctuation-checkpoints", filename="inference.py")

# 2. Load the script
spec = importlib.util.spec_from_file_location("inference", script_path)
inference = importlib.util.module_from_spec(spec)
spec.loader.exec_module(inference)

# 3. Initialize and Predict
model = inference.PunctuationRestorer()

text = "هل تساءلت يوما عن معنى الحياة ما هي الأسئلة التي تشغل بالك"
print(model.predict(text))
# Output: هل تساءلت يوماً عن معنى الحياة؟ ما هي الأسئلة التي تشغل بالك؟