CyberGPT-Qwen3-4B

CyberGPT is a cybersecurity-focused language model based on Qwen3-4B-Instruct-2507, fine-tuned using QLoRA/LoRA for cybersecurity-oriented question answering and security analysis.

The trained model was merged, quantized to Q4_K_M GGUF, and deployed locally using llama.cpp with CUDA acceleration.

Model Details

Property	Value
Base Model	Qwen3-4B-Instruct-2507
Fine-tuning	QLoRA / LoRA
LoRA Rank	16
LoRA Alpha	16
Quantization	Q4_K_M
Format	GGUF
Model Size	~2.3 GB
Inference Engine	llama.cpp
GPU Acceleration	NVIDIA CUDA
Deployment	llama-server

Fine-Tuning

CyberGPT was fine-tuned using parameter-efficient fine-tuning with QLoRA.

The LoRA configuration used:

• Rank: 16
• Alpha: 16
• Target modules:
  • q_proj
  • k_proj
  • v_proj
  • o_proj
  • gate_proj
  • up_proj
  • down_proj

The fine-tuned adapter was merged with the base model before quantization.

Quantization

The merged model was converted to GGUF and quantized using Q4_K_M.

This reduces the model size substantially while retaining useful language-model capability, making local inference practical on consumer GPUs.

Final model size: approximately 2.3 GB.

Local Inference

CyberGPT can be run locally using llama.cpp.

Example:

llama-cli -m CyberGPT-Q4_K_M.gguf -ngl 35 -c 2048