Yusiko/DOST_dataset

🇦🇿 Azerbaijani Friendly Instruct Dataset (Alpaca Format)

Welcome to the Azerbaijani Friendly Instruct Dataset! 👋

This dataset is designed to train Large Language Models (LLMs) to speak Azerbaijani not just correctly, but with soul, empathy, and local flavor. Unlike standard formal datasets, this collection aims to create an AI persona that feels like a close friend — supportive, slightly informal, and culturally aware.

🌟 Why this Dataset?

Most datasets focus on formal wikipedia-style QA. This dataset is different. It focuses on:

• ❤️ Empathy & EQ: The model is trained to offer psychological support, motivation, and understanding.
• 🗣️ Natural Speech: Uses local Azerbaijani interjections (e.g., "Uyda", "Can", "Vay vay") and informal sentence structures.
• 🤖 Tech & Science: Includes specific instructions on Mechatronics, Robotics, C++, Python, and Engineering, explained in simple terms.
• 🏙️ Local Culture: Knows about Azerbaijan life, national cuisine (Dolma, Plov), and local habits.

📂 Dataset Structure

The dataset follows the standard Alpaca format (JSON), making it compatible with most fine-tuning scripts (Axolotl, LLaMA-Factory, etc.).

Column	Description
instruction	The task or question provided by the user.
input	Optional context or additional information (can be empty).
output	The friendly, helpful, and "human-like" response from the AI.

📝 Example Entry

{
  "instruction": "I feel very tired and unmotivated today.",
  "input": "",
  "output": "Uyda, don't say that, my heart aches! 😔 I know, sometimes everything piles up at once. But this is just one day, not your whole life. Maybe lie down and rest a bit? Imagine I'm pouring you a hot cup of tea. ☕ You are strong!"
}

🎯 Categories

The dataset covers a wide range of topics:

• 🫂 Psychological Support: Dealing with stress, anxiety, and daily struggles.
• 💻 Coding & Tech: Python, C++, Arduino, Raspberry Pi, and troubleshooting.
• 🍳 Daily Life: Cooking, relationship advice, movies, and music.
• 🇦🇿 Cultural Context: Azerbaijani lifestyle, holidays (Novruz), and traditions.

⚠️ Limitations & Bias

• Informal Tone: This dataset is designed for chat/assistants. It may not be suitable for formal document generation (legal/medical).
• Synthetic Generation: While curated, some data is synthetically generated/augmented and should be verified for absolute factual accuracy.

🤝 Contribution

Contributions are welcome! If you want to add more "Friendly flavor" or technical instructions, feel free to open a Pull Request.

Let's build the friendliest AI for the Azerbaijani language! 🚀🇦🇿