nassimjp/problem_solving-reasoning-pashto-plus

Problem Solving & Reasoning Multilingual Dataset

This repository contains a specialized dataset focused on logical reasoning, problem-solving, and step-by-step cognitive workflows across multiple regional languages: Pashto (ps), Arabic (ar), Farsi (fa), Sindhi (sd), and Urdu (ur).

Dataset Overview

• —Languages: Pashto (پښتو), Arabic (العربية), Farsi (فارسی), Sindhi (سنڌي), Urdu (اردو)
• —Domain: Logical Reasoning, Problem Solving, Cognitive SFT
• —License: MIT

Structure

The dataset is structured to facilitate Supervised Fine-Tuning (SFT) and alignment for Large Language Models (LLMs) to enhance their reasoning capabilities.

{
  "instruction": "...",
  "input": "...",
  "output": "..."
}

Dataset Files

The data files are organized in the data/ directory by language:

• —data/pashto_chunk_01.jsonl
• —data/arabic_chunk_01.jsonl
• —data/farsi_chunk_01.jsonl
• —data/sindhi_chunk_01.jsonl
• —data/urdu_chunk_01.jsonl

Usage

You can easily load a specific configuration or data file using the Hugging Face datasets library:

from datasets import load_dataset

# To load the Pashto data file example:
dataset = load_dataset(
    "nassimjp/problem_solving-reasoning-pashto", 
    data_files="data/pashto_chunk_01.jsonl"
)
print(dataset["train"][0])

License

This project is released under the MIT License.