README.md

---
license: apache-2.0
task_categories:
- fill-mask
size_categories:
- 1K<n<10K
---

# Bangla Quran Punctuation Dataset

A high-quality dataset for Bangla punctuation restoration, derived exclusively from the Bangla translation of the Holy Quran.

## Dataset Description

This dataset is designed for training models on **punctuation restoration** in Bangla text. Each sample consists of:

- **human**: Bangla text with all punctuation removed
- **gpt**: The original Bangla text with correct punctuation (। , ; : - ! ?)

All samples are derived from the Bangla translation of the Quran (source: [zusmani/the-holy-quran](https://www.kaggle.com/datasets/zusmani/the-holy-quran)).

Long verses were intelligently split using a smart chunking algorithm that:
- Cuts only at sentence boundaries (। ! ?)
- Ensures every sample ≤ 256 characters
- Preserves complete sentences and all original content

**No data loss** — every word from the Quran is included.

## Statistics

- Total samples: 6846 (after smart chunking)
- Max length: 256 characters
- Min length: >20 characters
- Average length: ~130 characters
- All samples contain at least one punctuation mark


## Example

```json
{
  "conversations": [
    {
      "from": "human",
      "value": "আল্লাহর নামে শুরু করছি যিনি পরম করুণাময় অতি দয়ালু"
    },
    {
      "from": "gpt",
      "value": "আল্লাহর নামে শুরু করছি যিনি পরম করুণাময়, অতি দয়ালু।"
    }
  ],
  "source": "holy_quran_bangla",
  "score": 10.0
}
```

## Usage

Ideal for fine-tuning models on:
- Punctuation restoration
- Sentence boundary detection
- Bangla NLP in low-resource settings

```
@dataset{bangla_quran_punctuation_2025,
  title = {Bangla Quran Punctuation Restoration Dataset},
  author = {H.M. MEHEDI HASAN(Badhon)},
  year = {2025},
  publisher = {Hugging Face},
  version = {1.0},
  url = {https://huggingface.co/datasets/Badhon/BanglaQuranPunctuationDataset}
}
```