Perfect — you’re asking for different types of dataset examples for LLM / NLP (especially text-to-text, reasoning, code, chat). I’ll give you a complete structured set 🔥 so you can directly use them.

---

🧠 1. Basic Text-to-Text (Simple NLP)

👉 Direct input → output

{"input": "Translate Hello to Hindi", "output": "नमस्ते"}
{"input": "Summarize: AI is growing very fast.", "output": "AI is growing quickly."}

✔️ Use for:

• translation
• summarization

---

🧠 2. Instruction-Based Dataset (Most Important 🔥)

{"instruction": "Explain concept", "input": "SQL Injection", "output": "SQL Injection is a vulnerability where attackers inject malicious queries."}
{"instruction": "Define term", "input": "Machine Learning", "output": "Machine learning is a method where systems learn from data."}

✔️ Used in:

• modern LLMs
• instruction tuning

---

🧠 3. Question Answering (QA)

{"question": "What is Python?", "answer": "Python is a programming language."}
{"question": "What is NLP?", "answer": "Natural Language Processing is a field of AI."}

✔️ Simple but powerful

---

🧠 4. Conversational (ChatGPT Style 🔥)

{
  "messages": [
    {"role": "user", "content": "What is SQL Injection?"},
    {"role": "assistant", "content": "SQL Injection is a vulnerability..."}
  ]
}

---

🧠 5. Multi-turn Conversation (Advanced Chat)

{
  "messages": [
    {"role": "user", "content": "What is XSS?"},
    {"role": "assistant", "content": "XSS is a web vulnerability."},
    {"role": "user", "content": "Give example"},
    {"role": "assistant", "content": "<script>alert(1)</script>"}
  ]
}

---

🧠 6. Code Generation Dataset 🔥

{"instruction": "Write Python code for factorial", "output": "def fact(n): return 1 if n==0 else n*fact(n-1)"}
{"instruction": "Write JavaScript fetch API example", "output": "fetch('url').then(res=>res.json())"}

---

🧠 7. Reasoning (Chain-of-Thought) 🔥

{
  "instruction": "Check if number is prime",
  "output": "Step 1: Check divisibility\nStep 2: Loop from 2 to n\nStep 3: If divisible return false\nFinal code: def is_prime(n): ..."
}