Update README.md
Browse files
README.md
CHANGED
|
@@ -35,3 +35,11 @@ configs:
|
|
| 35 |
- split: train
|
| 36 |
path: data/train-*
|
| 37 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
- split: train
|
| 36 |
path: data/train-*
|
| 37 |
---
|
| 38 |
+
This dataset is an Arabic sample extracted from the [Fineeb2](https://huggingface.co/datasets/HuggingFaceFW/fineweb-2)
|
| 39 |
+
Arabic subset (arb_Arab) which is supposed to be standard Arabic.
|
| 40 |
+
There are around 2.3 million rows in this sample. First, the whole dataset (57.8M rows) was scanned and rows
|
| 41 |
+
were kept if they had over 95% Arabic words. Then this 2.3M sample was randomly sampled from the _mostly Arabic_ data.
|
| 42 |
+
Notice that language_score is not an accurate measure. Also, this did not exclude slang, dialects or inappropriate
|
| 43 |
+
content (no editing was done to any row and all columns were kept).
|
| 44 |
+
The main purpose of this dataset is educational and I hope it helps researchers in designing and developing pre-processing
|
| 45 |
+
for the main FineWeb2 dataset (or any other Arabic corpora).
|