GGUF quantized versions of AYI-NEDJIMI/CyberSec-Assistant-3B for use with Ollama, llama.cpp, LM Studio, and other GGUF-compatible inference engines.

Model Description

This is a fine-tuned Qwen2.5-3B-Instruct model specialized in general cybersecurity. It can answer questions about network security, vulnerability assessment, incident response, penetration testing, threat analysis, security architecture, and cybersecurity best practices in both French and English.

Part of the AYI-NEDJIMI Cybersecurity AI Portfolio:

AYI-NEDJIMI/CyberSec-AI-Portfolio — Full collection

Available Quantizations

Filename	Quant Type	Size	Description
`cybersec-assistant-3b-Q4_K_M.gguf`	Q4_K_M	1.80 GB	Recommended — Best balance of quality and size (~31% of F16)
`cybersec-assistant-3b-Q5_K_M.gguf`	Q5_K_M	2.07 GB	Higher quality, slightly larger (~36% of F16)
`cybersec-assistant-3b-Q8_0.gguf`	Q8_0	3.06 GB	Near-lossless quantization (~53% of F16)

Quantization Format Details

Q4_K_M: 4-bit quantization with k-quant medium quality. Excellent for resource-constrained environments. Minimal quality loss for most tasks.
Q5_K_M: 5-bit quantization with k-quant medium quality. Good middle ground between Q4 and Q8.
Q8_0: 8-bit quantization. Near-original quality with ~50% size reduction from F16.

How to Use

Ollama

Create a Modelfile:

FROM ./cybersec-assistant-3b-Q4_K_M.gguf

TEMPLATE """<|im_start|>system
{{ .System }}<|im_end|>
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""

SYSTEM "You are a cybersecurity expert assistant. You provide detailed, accurate guidance on network security, vulnerability assessment, incident response, penetration testing, and security best practices. You respond in the same language as the user's question."

PARAMETER temperature 0.7
PARAMETER top_p 0.8
PARAMETER top_k 20
PARAMETER stop "<|im_end|>"

Then run:

bash

ollama create cybersec-assistant -f Modelfile
ollama run cybersec-assistant

llama.cpp

bash

# Interactive chat
./llama-cli -m cybersec-assistant-3b-Q4_K_M.gguf \
  -p "You are a cybersecurity expert assistant." \
  --chat-template chatml \
  -cnv

# Server mode
./llama-server -m cybersec-assistant-3b-Q4_K_M.gguf \
  --host 0.0.0.0 --port 8080

LM Studio

Download the desired GGUF file
Open LM Studio and load the model from your downloads
Select the ChatML chat template
Set the system prompt to: "You are a cybersecurity expert assistant."
Start chatting!

Python (llama-cpp-python)

python

from llama_cpp import Llama

llm = Llama(model_path="cybersec-assistant-3b-Q4_K_M.gguf", n_ctx=4096)

response = llm.create_chat_completion(
    messages=[
        {"role": "system", "content": "You are a cybersecurity expert assistant."},
        {"role": "user", "content": "Explain the MITRE ATT&CK framework and how it helps in threat detection."}
    ],
    temperature=0.7,
    top_p=0.8,
    top_k=20,
)
print(response["choices"][0]["message"]["content"])

Related Models

Version	Link
Merged (SafeTensors)	AYI-NEDJIMI/CyberSec-Assistant-3B
LoRA Adapter	AYI-NEDJIMI/CyberSec-Assistant-3B-Adapter
GGUF (this repo)	AYI-NEDJIMI/CyberSec-Assistant-3B-GGUF
Portfolio Collection	AYI-NEDJIMI/CyberSec-AI-Portfolio

Technical Details

Base Model: Qwen/Qwen2.5-3B-Instruct
Fine-tuning: QLoRA (4-bit) with LoRA adapters merged back
Architecture: Qwen2ForCausalLM
Context Length: 4096 tokens
Chat Template: ChatML
Converted with: llama.cpp (convert_hf_to_gguf.py)

AYI-NEDJIMI/CyberSec-Assistant-3B-GGUF

CyberSec-Assistant-3B-GGUF