rephrase_mixed

rephrase_mixed is a large-scale Arabic text rephrasing dataset generated using Gemma 4 E4B from content in AdaMLLab/AraMix-HQ.

The dataset is intended for research and development involving Arabic text generation, paraphrasing, and synthetic-data-based language-model training.

Dataset summary

Property	Value
Dataset repository	Zaid/rephrase_mixed
Parent dataset	AdaMLLab/AraMix-HQ
Generation model	Gemma 4 E4B
Language	Arabic
Split	train
Number of records	19,939,441
Number of tokens	~ 37 Billion tokens
License	Same as AraMix-HQ: see individual source licenses; non-commercial use
Repository data format	Parquet
Record fields	id, num_output_tokens, text

1. Parent dataset and source selection

The parent dataset is AdaMLLab/AraMix-HQ.

AraMix-HQ is a quality-filtered Arabic corpus derived from AraMix-MinHash. Its published description identifies multiple upstream sources, including CulturaX, ArabicWeb24, HPLT 2.0, FineWeb-2, C4, ClusterLab 101B, and FinePDFs. AraMix-HQ does not assign one blanket license to all of these sources; it directs users to the individual source-dataset licenses.

Source coverage and selection procedure for this release:

The released train split contains 19,939,441 generated records. This count describes the current rephrased dataset; We only processed a subset containing the first 77 shard files.

2. Dataset contents and structure

Each record contains the following fields:

Field	Type	Description
id	string	Identifier associated with the source/document record.
num_output_tokens	integer	Number of tokens in the generated rephrased output, according to the tokenizer used during dataset creation.
text	string	Arabic rephrased text generated by Gemma 4 E4B.

The text field contains the generated content. The available dataset preview shows structured text that may include headings, tables, and question-and-answer-style organization. These structures are represented as text (including Markdown-style formatting) within the text field, rather than as separate table or question/answer columns.

The dataset is distributed in Parquet format on Hugging Face. It has one split, train.

Example

{
  "id": "urn:uid:...",
  "num_output_tokens": 1960,
  "text": "..."
}

3. Generation

The rephrased records were generated using Gemma 4 E4B.

Generation prompt(s):

Rewrite the given document in {{ language }} as the following styles:
  1. A structured table that organizes the key information, then generate one "
  "question-answer pair based on the table. First extract the main data points and organize them into a clear "
  "table format with appropriate headers using markdown table syntax with proper alignment. After the table, "
  "generate one insightful question that can be answered using the table data. Provide a clear, concise answer "
  "to the question based on the information in the table. Output only the table followed by the question-answer "
  "pair.
  2. A comprehensive FAQ (Frequently Asked Questions). Extract or infer the key questions "
  "a reader would have about this topic, then provide clear, direct answers. Order questions logically, from "
  "foundational to advanced, or by topic area. Each answer should be self-contained and understandable without "
  "reference to other answers. Ensure the FAQ works as a standalone document.
  3. A clear, step-by-step tutorial or instructional guide. Use numbered steps or bullet "
  "points where appropriate to enhance clarity. Preserve all essential information while ensuring the style "
  "feels didactic and easy to follow.
  4. A mathematical word problem based on the numerical data or relationships in "
      "the text. Provide a step-by-step solution that shows the calculation process clearly. Create a problem that "
      "requires multi-step reasoning and basic arithmetic operations. It should include the question followed by a "
      "detailed solution showing each calculation step. 
  Document: [xxxx]

4. License and terms of use

Same as AraMix-HQ: see individual source licenses; non-commercial use.

7. Citation

If you use rephrase_mixed, please cite the dataset repository and the upstream AraMix work as appropriate.

AraMix reference:

@misc{alrashed2025mixminmatch,
  title        = {Mix, MinHash, and Match: Cross-Source Agreement for Multilingual Pretraining Datasets},
  author       = {Sultan Alrashed and Francesco Orabona},
  year         = {2025},
  eprint       = {2512.18834},
  archivePrefix = {arXiv},
  primaryClass = {cs.CL},
  url          = {https://arxiv.org/abs/2512.18834}
}

---