Cyber Security Instruction Dataset

Dataset Summary

Cyber Security Instruction Dataset is an instruction-following dataset created for fine-tuning Large Language Models (LLMs) in cybersecurity and penetration testing tasks.

The dataset focuses on high-quality question-answer pairs covering defensive security, ethical hacking, secure coding, AI security, and vulnerability assessment.

---

Features

• Instruction tuning format
• Multi-turn ready
• Human-readable responses
• LLM fine-tuning compatible
• PEFT / LoRA compatible
• Hugging Face Datasets compatible

---

Supported Topics

• Web Application Security
• Network Security
• API Security
• Mobile Security
• Cloud Security
• AI Security
• LLM Security
• Prompt Injection
• Jailbreak Attacks
• RAG Security
• Secure Coding
• Python Security
• Linux Security
• Windows Security
• Docker Security
• Kubernetes Security
• DevSecOps
• OWASP Top 10
• OWASP API Top 10
• MITRE ATT&CK
• MITRE ATLAS
• CVE Analysis
• CVSS
• Burp Suite
• Nmap
• SQLMap
• Metasploit
• Wireshark
• Nuclei
• Reconnaissance
• VAPT Reporting

---

Dataset Format

Each record follows the instruction tuning format.

Example:

{
  "instruction": "What is SQL Injection?",
  "input": "",
  "output": "SQL Injection is a vulnerability that allows attackers to manipulate SQL queries by injecting malicious input..."
}

Chat format example:

{
  "messages": [
    {
      "role": "user",
      "content": "Explain Cross Site Scripting."
    },
    {
      "role": "assistant",
      "content": "Cross Site Scripting (XSS) allows attackers to inject malicious JavaScript into web pages..."
    }
  ]
}

---

Intended Use

This dataset is intended for:

• Supervised Fine-Tuning (SFT)
• LoRA Training
• QLoRA Training
• DPO
• ORPO
• Instruction Tuning
• Research
• Educational Purposes

---

Data Sources

The dataset is curated from:

• Public cybersecurity documentation
• Official security standards
• Open educational resources
• Public vulnerability documentation
• Original human-written instruction-response pairs

---

Quality Control

The dataset is manually reviewed for:

• Formatting consistency
• Duplicate removal
• Instruction clarity
• Grammar
• Response quality
• Safety

---

Limitations

• May not contain the latest vulnerabilities.
• Should not replace official vendor documentation.
• Responses may require updates as cybersecurity evolves.

---

License

Apache License 2.0

---

Citation

If you use this dataset in your research or project, please cite this repository.

---

Maintainer

Nitin Saini

GitHub: https://github.com/

Hugging Face: https://huggingface.co/Nitinsaini077