Update README.md
Browse files
README.md
CHANGED
|
@@ -25,7 +25,7 @@ default: minhash_deduped
|
|
| 25 |
|
| 26 |
AraMix ([https://arxiv.org/abs/2512.18834](https://arxiv.org/abs/2512.18834)) is an Arabic pretraining corpus containing 178 billion tokens across 179 million documents (in the minhash subset). Rather than scraping the web again, AraMix combines seven publicly available Arabic datasets, applies Arabic-specific quality filtering, and performs cross-dataset deduplication.
|
| 27 |
|
| 28 |
-
We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that the `matched` subset of AraMix, outperforms the previous state-of-the-art model-free approach, [arabicweb24](https://huggingface.co/datasets/lightonai/ArabicWeb24) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)) and the previous state-of-the-art model-based approach, [FineWeb2-HQ](https://huggingface.co/datasets/epfml/FineWeb2-HQ) while having more total tokens. Furthermore, the `minhash_deduped` subset performs on-par with nearly 5 times the total number of tokens.
|
| 29 |
|
| 30 |
## Subsets
|
| 31 |
|
|
|
|
| 25 |
|
| 26 |
AraMix ([https://arxiv.org/abs/2512.18834](https://arxiv.org/abs/2512.18834)) is an Arabic pretraining corpus containing 178 billion tokens across 179 million documents (in the minhash subset). Rather than scraping the web again, AraMix combines seven publicly available Arabic datasets, applies Arabic-specific quality filtering, and performs cross-dataset deduplication.
|
| 27 |
|
| 28 |
+
We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that the `matched` subset of AraMix, outperforms the previous state-of-the-art model-free approach, [arabicweb24](https://huggingface.co/datasets/lightonai/ArabicWeb24) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)) and the previous state-of-the-art model-based approach, [FineWeb2-HQ](https://huggingface.co/datasets/epfml/FineWeb2-HQ), all while having more total tokens and being completely model-free. Furthermore, the `minhash_deduped` subset performs on-par with nearly 5 times the total number of tokens.
|
| 29 |
|
| 30 |
## Subsets
|
| 31 |
|