Polygl0t/hindi-edu-qwen-annotations

Annotations for the Hindi Edu classifier 📚

Supported Tasks and Leaderboards

This dataset can be used for the task of text classification, specifically for educational quality assessment in Hindi text.

Languages

Hindi.

Dataset Structure

Data Instances

• id: a unique identifier for each sample (md5 hash).
• text: a string of text in Hindi.
• source: the source where that string originated.
• subset: a short string indicating the name of the subset (referring to the original dataset or crawl).
• score: the score assigned by Qwen/Qwen2.5-32B-Instruct.

Data Fields

{
  "text": "अध्ययनकर्ताओं ने पाया है कि एपीसी5 मिट्टी में पाए जाने वाले हानिकारक पॉलीसाइक्लिक एरोमेटिक हाइड्रोकार्बन (पीएएच) जैसे कार्बनिक अवशिष्ट पदार्थों को अपघटित कर सकता है।",
  "source": "https://huggingface.co/datasets/HuggingFaceFW/fineweb-2",
  "subset": "finweb_2_hi",
  "id": "e9162717a61b5b1cb2dc83fd5419dde9",
  "edu_score": 4
}

Subsets and Splits

This dataset contains a single split: train, which includes all 400,000 samples.

from datasets import load_dataset

# Load the main dataset
ds = load_dataset("Polygl0t/hindi-edu-qwen-annotations", split="train")

# If you don't want to download the entire dataset, set streaming to `True`
ds = load_dataset("Polygl0t/hindi-edu-qwen-annotations", split="train", streaming=True)

Dataset Creation

Source Data

All data was sourced from Polygl0t/gigalekh-v1. The id field corresponds to the md5 hash of the text.

Annotations

The dataset was created to provide high-quality educational annotations for Hindi text, leveraging the capabilities of Qwen2.5 to generate consistent and scalable annotations.

The score distribution of this dataset is the following:

Edu Score	N samples
1	206822
2	135624
3	46287
4	11214
5	53

Annotation Process

The initial annotation was performed on a random sample of 400,000 documents from Polygl0t/gigalekh-v1, where we tried to sample equally from all the subsets that form this corpus. In terms of infrastructure, we used 4 x NVIDIA A100-SXM4-80GB GPUs and the vLLM library as our inference engine with 4-fold tensor parallelism.

The prompt used to make Qwen2.5 work as an annotator for educational data is the following:

You are a text quality evaluator for the Hindi language. Your task is to assess, from a scale of 1 to 5, how informative and educational a given text sample is. You MUST output your score in a JSON format.

### Evaluation Criteria:
- Give 1 if the text is not informative or educational. Also, give 1 if the text is too simple, too short, ill-formatted, non-sensical, or contains NSFW content.
- Give 2 if the text is somewhat informative but lacks educational value. For example, it might mix educational content with non-educational material, offering a superficial overview of potentially useful topics.
- Give 3 if the text is informative and appropriate for educational use and introduces key concepts relevant to school curricula. For example, if the text is a well-written article on a scientific topic, but it it may not be comprehensive or could include some extraneous information, being overly complex or too simple.
- Give 4 if the text is educational and informative, providing a highly relevant and beneficial for educational purposes for a level not higher than grade school, exhibiting a clear and consistent writing style. For example, it could be similar to a chapter from a textbook or a tutorial, offering substantial educational content, including exercises and solutions, with minimal irrelevant information.
- Give 5 if the text is highly educational and informative. For a 5 score, the text must be outstanding in its educational value, perfectly suited for teaching either at primary school or grade school. It follows detailed reasoning, the writing style is easy to follow and offers profound and thorough insights into the subject matter.

### Output Requirements:
- Your output must be a valid JSON object.
- The JSON must contain only one key-value pair.
- The key must be 'score'.
- The value must be an integer between 1 and 5.
- The value must be a faithful representation of the criteria above.

### Example Output:
\`\`\`json
{'score': 2}
\`\`\`

Ensure that your responses strictly adhere to this format and evaluate the text accurately based on the given criteria.

Considerations for Using the Data

Social Impact of Dataset

The dataset aims to provide a valuable resource for understanding and improving educational content in Hindi text. By offering high-quality annotations, it can help researchers and developers create more effective educational tools, promote better learning outcomes, and contribute to the broader field of NLP for low-resource languages.

Additional Information

Dataset Maintainers

• Shiza Fatimah.
• Nicholas Kluge Corrêa.
• Aniket Sen.

Licensing Information

The dataset is licensed under the Apache-2.0 License.

Citation Information

@misc{shiza2026lilmoo,
      title={{Raising Bars, Not Parameters: LilMoo Compact Language Model for Hindi}}, 
      author={Shiza Fatimah and Aniket Sen and Sophia Falk and Florian Mai and Lucie Flek and Nicholas Kluge Corr{\^e}a},
      year={2026},
      eprint={2603.03508},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2603.03508}, 
}

Acknowledgments

Polyglot is a project funded by the Federal Ministry of Education and Research (BMBF) and the Ministry of Culture and Science of the State of North Rhine-Westphalia (MWK) as part of TRA Sustainable Futures (University of Bonn) and the Excellence Strategy of the federal and state governments.

We also gratefully acknowledge the granted access to the Marvin cluster hosted by University of Bonn along with the support provided by its High Performance Computing & Analytics Lab.

Contributions

If you want to contribute, contact us at polyglot@uni-bonn.de!