I fine-tuned a 2B model for Cybersecurity & CTF triage that runs completely offline on an 8GB laptop (Ollama + WebUI)
retained text
Hey everyone,
Most specialized cybersecurity LLMs out there are massive 70B+ beasts that require dual A100s, or they're gated behind cloud APIs where you can't paste sensitive logs, payloads, or client code due to confidentiality.
I wanted something lightweight, private, and air-gapped that could run on a regular consumer laptop without cooking the CPU.
So I fine-tuned **CyberQwen-2B** — a domain-specific model optimized for quick security triage, explaining attack vectors, decoding obfuscated payloads, and helping with CTF challenges.
# ⚡ Highlights:
* Hardware-friendly: Runs at \~12–15 tokens/sec on an older 8th-gen Intel i5 laptop with 8GB RAM (CPU-only).
* 100% Offline & Private: Zero data leaves your machine. Perfect for bug bounty recon, internal network logs, and sensitive source code.
* Native Ollama integration: Easy to pull and run via a single Modelfile.
* Built-in WebUI & Mobile Access: Includes a zero-dependency streaming WebUI + an automatic Cloudflare tunnel script if you want to access the model from your phone while on the go (no cloud hosting fees).
# 🛠️ What it's trained for:
* Vulnerability Triage & CVE Context: Explaining CVE mechanisms, impact, and remediation steps.
* Payload & Shellcode Deconstruction: Breaking down SQLi, XSS, and obfuscated PowerShell/Bash commands.
* CTF & Reverse Engineering Assistance: Analyzing decompiled assembly/C snippets and suggesting exploitation strategies.
# 🚀 How to Run It:
Everything is open-source. You can run it with Ollama or the included CLI triage tool:
1. Clone the repo:
git clone https://github.com/sambuddha001/cyberqwen-2b.git
2. Run setup:
bash setup.sh
3. Chat:
ollama run cybersec-qwen
GitHub: [https://github.com/sambuddha001/cyberqwen-2b](https://github.com/sambuddha001/cyberqwen-2b)
I'd love feedback from the community—especially on edge cases, prompts where it struggles, or ideas for benchmark improvements!document fetched 2026-11-15 00:36 · canonical https://reddit.com/r/ollama/comments/1xw84q1/i_finetuned_a_2b_model_for_cybersecurity_ctf