---
title: Code Vulnerability Detector
emoji: 🔐
colorFrom: red
colorTo: pink
sdk: gradio
app_file: app.py
pinned: false
---

# 🔐 Code Vulnerability Detector

Deteksi celah keamanan dalam kode secara otomatis menggunakan AI. Model menganalisis kode dan memberikan laporan terstruktur berisi jenis vulnerability, tingkat bahaya, penjelasan, dan saran perbaikan.

## 🚀 Cara Pakai

1. Pilih bahasa pemrograman
2. Paste kode yang ingin dianalisis
3. Klik **Analyze Code**
4. Model akan generate laporan keamanan terstruktur

## 📋 Output Format

```
Vulnerability : [tipe vulnerability]
Severity      : [HIGH/MEDIUM/LOW]
Language      : [bahasa pemrograman]
Penjelasan    : [penjelasan kenapa berbahaya]
Fix           : [kode yang sudah diperbaiki]
```

## 🛠️ Model Details

| Parameter | Value |
|-----------|-------|
| Base Model | Qwen/Qwen2.5-Coder-0.5B-Instruct |
| Method | LoRA fine-tuning |
| Framework | MLX-LM (Apple Silicon) |
| Dataset | CyberNative/Code_Vulnerability_Security_DPO |
| Train Samples | 4,187 |
| Training Steps | 1,000 |
| Learning Rate | 2e-4 |
| Accuracy | 62% |

## 🌐 Bahasa yang Didukung

Python, JavaScript, Java, C++, C#, PHP, Ruby, Swift, Go, Kotlin, Fortran

## 🔍 Jenis Vulnerability yang Dideteksi

- SQL Injection
- Cross-Site Scripting (XSS)
- Buffer Overflow
- Command Injection
- Improper Authentication
- Insecure Deserialization
- Dan banyak lagi...

## 📊 Dataset

Dataset dari **[CyberNative/Code_Vulnerability_Security_DPO](https://huggingface.co/datasets/CyberNative/Code_Vulnerability_Security_DPO)**:
- 4,656 pasangan kode vulnerable vs secure
- 11 bahasa pemrograman
- Format DPO (Direct Preference Optimization)

## 👤 Author

**Bernadetta**  
[HuggingFace](https://huggingface.co/Bernadetta14)