nassimjp/pashto-qwen-1k-cot

Pashto-Qwen-1K-CoT Dataset

Overview

Pashto-Qwen-1K-CoT is a high-quality reasoning dataset consisting of 1,024 samples, specifically curated to enhance the Chain-of-Thought (CoT) capabilities of Pashto language models. This dataset is a translated version of a subset from brendan-gho/qwen3b_paraphrased_cat_cot.

By focusing on "Reasoning" rather than just "Information," this dataset helps models like Baran and Roshan develop logical thinking paths in the Pashto language.

Dataset Structure

The dataset is provided in CSV format with the following schema:

• —prompt_en: The original complex reasoning prompt (English).
• —prompt_ps: The translated reasoning prompt (Pashto).
• —completion_en: The step-by-step logic and final answer (English).
• —completion_ps: The translated step-by-step logic and final answer (Pashto).

Why This Dataset?

Unlike traditional web-scraped data, this dataset provides:

1. 1.Logical Depth: It forces the model to use <think> tags to break down problems before providing an <answer>.
2. 2.Proven Quality: Based on the paraphrased outputs of the Qwen-3B model, ensuring high-quality reasoning chains.
3. 3.iPashto.ai Standards: Translated and post-processed to align with the linguistic standards of the iPashto.ai initiative.

Methodology

• —Source: Paraphrased Cat CoT (Qwen-3B based).
• —Translation: Automated via Google Translate API with manual oversight for tag preservation.
• —Cleaning: Applied custom regex filters to fix broken XML tags (e.g., converting translated "فکر وکړئ" back to </think>).

About iPashto.ai

This dataset is a contribution from nassimjp (نصیب الله), an IT Specialist and System Architect based in Japan. Through the iPashto.ai project, we are building the foundational infrastructure for the next generation of Pashto AI.

• —Founder: Nassim
• —Location: Saitama, Japan
• —Project Goal: Digital preservation and AI advancement for the Pashto language.

License

Licensed under Apache-2.0.