README.md

---
license: mit
datasets:
- Mireu-Lab/NSL-KDD
metrics:
- accuracy
- f1
base_model: Qwen/Qwen2.5-1.5B
pipeline_tag: question-answering
tags:
- intrusion-detection
- cybersecurity
- llm
- qwen
- tabular-reasoning
---
# PCF_ID: A Novel Prompt Cast Framework for Intrusion Detection
[![Hugging Face Model](https://img.shields.io/badge/Hugging%20Face-Model-blue)](https://huggingface.co/PHZane/PCF_ID-0.5B)
[![Python 3.8+](https://img.shields.io/badge/Python-3.8%2B-blue)](https://www.python.org/)
[![License](https://img.shields.io/badge/License-MIT-green.svg)](https://opensource.org/licenses/MIT)
## 📌 Model Overview
**PCF_ID-1.5B** is a fine-tuned version of **Qwen2.5-1.5B**, adapted for network intrusion detection using the **Prompt Cast Framework (PCF-ID)**. By transforming structured tabular records into semantically rich **Question–Reasoning–Answer (Q-R-A)** prompts, PCF_ID enables large language models to effectively reason over cybersecurity data.


PCF_ID-1.5B significantly outperforms:
- Raw LLMs (e.g., vanilla Qwen2.5-1.5B),
- Traditional machine learning models (e.g., Random Forest, SVM),
- Advanced graph neural networks (GNNs).
## 💻 Code & Reproduction
The full framework, preprocessing pipeline, and evaluation scripts are available at:  
🔗 [https://github.com/Zaneph1/PCF_ID](https://github.com/Zaneph1/PCF_ID)