Datasets:

Modalities:
Text
Formats:
parquet
Languages:
Arabic
ArXiv:
License:
SultanR commited on
Commit
5169b42
·
verified ·
1 Parent(s): 2a0ea4d

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +1 -3
README.md CHANGED
@@ -17,12 +17,10 @@ language:
17
  - ar
18
  ---
19
 
20
- # AraMix
21
 
22
  AraMix (https://arxiv.org/abs/2512.18834v1) 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.
23
 
24
- <img src="https://huggingface.co/datasets/AdaMLLab/AraMix/resolve/main/finetasks_arabic_results.png" width="900" alt="Finetasks benchmark scores, showing AraMix-Consensus as SOTA.">
25
-
26
  We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that the `consensus` subset of AraMix outperforms the previous state-of-the-art, [arabicweb24](https://huggingface.co/datasets/lightonai/ArabicWeb24) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)) while having more total tokens. Furthermore, the `minhash_deduped` subset performs on-par with nearly 5 times the total number of tokens.
27
 
28
  In this setup, we remove all samples in `consensus` with more than 5 duplicates.
 
17
  - ar
18
  ---
19
 
20
+ <img src="https://huggingface.co/datasets/AdaMLLab/AraMix/resolve/main/finetasks_arabic_results.png" width="900" alt="Finetasks benchmark scores, showing AraMix-Consensus as SOTA.">
21
 
22
  AraMix (https://arxiv.org/abs/2512.18834v1) 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.
23
 
 
 
24
  We train a 1.4B parameter language model through nanotron on 30 billion tokens to show that the `consensus` subset of AraMix outperforms the previous state-of-the-art, [arabicweb24](https://huggingface.co/datasets/lightonai/ArabicWeb24) (see [Appendix A9 in the Fineweb-2 paper](https://arxiv.org/pdf/2506.20920)) while having more total tokens. Furthermore, the `minhash_deduped` subset performs on-par with nearly 5 times the total number of tokens.
25
 
26
  In this setup, we remove all samples in `consensus` with more than 5 duplicates.