lumees/wikipedia-turkish-synthetic-query

---
readme: "main"
language:
- tr
license: cc-by-sa-4.0
tags:
- custom-dataset
- information-retrieval
- synthetic
size_categories:
- 10K<n<100K
---

# Turkish Wikipedia Synthetic Query Dataset

## Dataset Description

This dataset consists of pairs containing a `query` and a `pos` (positive document/passage). It is processed from raw JSONL files, ensuring no missing values in key fields.

### Data Fields
- `query`: The search query or input text.
- `pos`: The positive label, relevant document, or passage.

## Dataset Creation & Methodology

### Source Data
The positive passages (`pos`) were collected from the **Finewiki** dataset, ensuring high-quality, information-rich text as the foundation for retrieval tasks.

### Synthetic Generation
The corresponding queries (`query`) were synthetically generated to match the positive passages.
- **Model:** Generation was performed using **unsloth/Qwen3-30B-A3B-Instruct-2507-GGUF**.
- **Compute:** The generation pipeline was executed on **NVIDIA L40** GPUs.

### Dataset State
This dataset contains the random 10K samples from the collected data. We are working on the rest to complete queries for all positives and version with hard negatives.

## Citation

If you use this dataset, please cite Lumees AI:

```bibtex
@misc{lumees2024hardnegatives,
author = {Hasan KURŞUN, Kerem Berkay YANIK},
title = {CodeSearchNet Hard Negatives (Filtered)},
year = {2025},
publisher = {Lumees AI},
howpublished = {\url{[https://lumees.io](https://lumees.io)}},
email = {[e-mail]}
}
```