Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Ali Data — Urdu & Roman Urdu Corpus
|
| 2 |
+
|
| 3 |
+
A large public-domain/open collection of **Urdu (اردو)** and **Roman Urdu** text for training Urdu language models.
|
| 4 |
+
|
| 5 |
+
## Stats
|
| 6 |
+
- **21,846,033 unique rows** (deduplicated by exact text)
|
| 7 |
+
- Format: JSON Lines — each line: `{"text": "...", "source": "..."}`
|
| 8 |
+
- Scripts: Urdu (Arabic script) + Roman Urdu (Latin script)
|
| 9 |
+
|
| 10 |
+
## Sources
|
| 11 |
+
| Source | Rows (approx) |
|
| 12 |
+
|---|---|
|
| 13 |
+
| Hugging Face open datasets (news, QA, sentiment, translation, poetry, instructions) | ~17.5M |
|
| 14 |
+
| FineWeb-2 `urd_Arab` (web crawl, Urdu script) | ~3.3M |
|
| 15 |
+
| GitHub open corpora (poetry, SMS, reviews) | ~1.4M |
|
| 16 |
+
| FineWeb-2 `urd_Latn` (web crawl, Roman Urdu) | ~70K |
|
| 17 |
+
| Kaggle open datasets (news, poetry, reviews) | ~23K |
|
| 18 |
+
| BBC Urdu articles + Internet Archive Urdu books | ~160 |
|
| 19 |
+
|
| 20 |
+
## Usage
|
| 21 |
+
```python
|
| 22 |
+
from datasets import load_dataset
|
| 23 |
+
ds = load_dataset("PAILLM/ali-data", split="train")
|
| 24 |
+
# or stream it:
|
| 25 |
+
ds = load_dataset("PAILLM/ali-data", split="train", streaming=True)
|
| 26 |
+
```
|
| 27 |
+
|
| 28 |
+
## Notes
|
| 29 |
+
- Raw public collection; clean/filter for your training needs (language ID, quality, toxicity) before use.
|
| 30 |
+
- Respects each source's original license; compiled for research use.
|