NiffyHunt90/codeguard-security-7b

CodeGuard Security 7B

LoRA adapter fine-tuned on Qwen 2.5 7B Instruct for code vulnerability detection. Trained on 32 security patterns across 8 vulnerability categories to identify and explain security flaws in source code.

Vulnerabilities Detected

Category	CWE	Severity
SQL Injection	CWE-89	Critical
Command Injection	CWE-78	Critical
Hardcoded Secrets	CWE-798	Critical
Insecure Deserialization	CWE-502	Critical
XML External Entity (XXE)	CWE-611	High
Path Traversal	CWE-22	High
Server-Side Request Forgery	CWE-918	High
Unsafe Deserialization	CWE-502	High

Dataset

Trained on curated code security examples from real-world vulnerability disclosures, bug bounty reports, and secure code review patterns. Covers OWASP Top 10, CWE Top 25, and SANS 25. No synthetic or GPT-generated data.

How to use

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

base_model = AutoModelForCausalLM.from_pretrained(
    "Qwen/Qwen2.5-7B-Instruct",
    torch_dtype=torch.float16,
    device_map="auto"
)
model = PeftModel.from_pretrained(base_model, "NiffyHunt90/codeguard-security-7b")
tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-7B-Instruct")

code = '''
query = "SELECT * FROM users WHERE id = " + user_input
cursor.execute(query)
'''
prompt = f"Analyze this code for security vulnerabilities:\n{code}"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

Training

• Base model: Qwen 2.5 7B Instruct
• Method: LoRA
• Adapter size: 154 MB
• Hardware: 2x Tesla T4 (14.5GB VRAM)
• Framework: Unsloth + HuggingFace TRL

Related models

• WraithWall Core V3 — full security operations model
• WraithCore 7B — lightweight 616MB security adapter

Author

Adewale Babalola (Niffyhunt) — Founder, WraithWall

• niffyhunt.online
• wraithwall.online