nikhilnebhavani/hinglish-llama3-dataset

🇮🇳 Hinglish Instruction Dataset for Llama 3.1 8B

A conversational Hinglish dataset built for fine-tuning Llama 3.1 8B.
Hinglish = Hindi + English mixed in Roman script — exactly how people type on WhatsApp, Instagram, and daily chat.

---

📌 Dataset Description

This dataset contains single-turn instruction-response pairs in Hinglish (Roman script only).
No Devanagari script — pure WhatsApp-style typing.

The goal is to make Llama 3.1 8B respond naturally in Hinglish, the way a helpful desi friend would reply.

---

📊 Dataset Details

Property	Value
Language	Hinglish (Hindi + English, Roman script)
Format	JSONL — Llama 3.1 Chat Template
Model Target	meta-llama/Meta-Llama-3.1-8B-Instruct
Task	Instruction Following / Conversational AI
Script	Roman only (no Devanagari)
Built With	AI-assisted (Claude) + Human curation

---

💬 System Prompt Used

Tu ek helpful aur friendly AI hai jo Hinglish mein baat karta hai — 
Roman script mein Hindi aur English mix karke, exactly jaise log WhatsApp 
pe type karte hain. Kabhi bhi Hindi script use mat kar.

---

🗂️ Format

Each row contains a text field with full Llama 3.1 chat template:

{
  "text": "<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n
  Tu ek helpful aur friendly AI hai...<|eot_id|>
  <|start_header_id|>user<|end_header_id|>\n
  Kya roz multivitamin lena safe hai?<|eot_id|>
  <|start_header_id|>assistant<|end_header_id|>\n
  Haan generally safe hai yaar...<|eot_id|>"
}

---

📝 Example Conversations

User	Assistant
Kya roz multivitamin lena safe hai?	Haan generally safe hai yaar, but balanced diet ho toh bahut zaroorat nahi padti...
Social media dekhne se mood kharab ho jaata hai, kya karu?	Screen time limit set kar apne phone mein, aur unfollow karo jo accounts negative feel karten...
Bukhar hai mujhe, doctor ke paas kab jaana chahiye?	100.4°F se upar ho ya 2-3 din se zyada rahe toh zaroor doctor ke paas jao yaar...

---

🔧 Compatible Fine-tuning Tools

Tool	Status
🦥 Unsloth	✅ Ready
🤗 TRL SFTTrainer	✅ Ready
Axolotl	✅ Ready
LLaMA-Factory	✅ Ready

---

🚀 How to Load

from datasets import load_dataset

dataset = load_dataset("your-username/hinglish-llama3-dataset")
print(dataset["train"][0])

---

⚠️ Intended Use

• ✅ Fine-tuning open-source LLMs for Hinglish
• ✅ Research on code-switching / multilingual NLP
• ✅ Building Hinglish chatbots for Indian users
• ❌ Not for harmful, abusive, or misleading applications

---

📜 License

CC BY 4.0 — Free to use with attribution.

---

🙏 Acknowledgements

• Dataset built with assistance from Claude (Anthropic)
• Inspired by real Hinglish conversations from everyday Indian internet culture
• Model target: Meta Llama 3.1

---

✍️ Citation

If you use this dataset, please credit:

@dataset{hinglish_llama3_2024,
  title     = {Hinglish Instruction Dataset for Llama 3.1},
  author    = {Nikhil Nebhavani},
  year      = {2025},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasets/nikhilnebhavani/hinglish-llama3-dataset}
}