---
language: code
tags:
- vulnerability-detection
- code-security
- codebert
- python
- CWE-89
- CWE-78
- CWE-79
- CWE-352
- CWE-94
- CWE-22
- CWE-601
datasets:
- VUDENC
---

# PyGuard V2 — Python Vulnerability Detector

## Model Description
PyGuard is a fine-tuned Microsoft CodeBERT model for detecting
security vulnerabilities in Python code.

## Training Data
Trained on VUDENC dataset by Wartschinski et al. 2022
(Information and Software Technology Journal)
DOI: 10.5281/zenodo.3559841

## Results vs VUDENC (Wartschinski et al. 2022)

| Metric    | VUDENC (LSTM) | PyGuard V2 (CodeBERT) |
|-----------|--------------|----------------------|
| Precision | 82-96%       | 100.00%              |
| Recall    | 78-87%       | 100.00%              |
| F1 Score  | 80-90%       | 100.00%              |
| Accuracy  | N/A          | 100.00%              |

## Vulnerabilities Detected
- CWE-89: SQL Injection
- CWE-78: Command Injection
- CWE-79: Cross-Site Scripting (XSS)
- CWE-352: CSRF
- CWE-94: Remote Code Execution
- CWE-22: Path Disclosure
- CWE-601: Open Redirect

## Usage
```python
from transformers import AutoTokenizer
import torch
import torch.nn as nn
from transformers import AutoModel

tokenizer = AutoTokenizer.from_pretrained("Umair011/pyguard-v2")
```

## Citation
If you use this model please cite VUDENC:
Wartschinski, L., et al. (2022). VUDENC: Vulnerability Detection
with Deep Learning on a Natural Codebase for Python.
Information and Software Technology, 144.