Modelpublic

boraoxkan/codereview-ai

sourceHugging Facemitupdated 10mo agoView on Hugging Face

0likes14downloads

Model Card

CodeReview AI

Automated Code Review with Fine-tuned LLMs

![GitHub](https://github.com/boraoxkan/CodeReview) ![License](https://opensource.org/licenses/MIT) ![Base Model](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct)

</div>

Overview

A fine-tuned code review model that automatically detects bugs, security vulnerabilities, and code quality issues across multiple programming languages.

Key Features

—Multi-Language: Python, JavaScript, Java, C++, Go, Rust, TypeScript, C#, SQL
—Security Focus: Detects OWASP Top 10 vulnerabilities
—Quality Scoring: 0-100 score with explanations
—Auto-Fix: Provides corrected code snippets
—Efficient: 4-bit quantization, runs on 8GB VRAM

Model Details

Property	Value
Base Model	Qwen2.5-Coder-7B-Instruct
Parameters	7B
Fine-tuning	LoRA (r=16, alpha=16)
Quantization	4-bit NF4
Context Length	2048 tokens
Framework	Unsloth + TRL

Detected Issues

Security

—SQL Injection
—Cross-Site Scripting (XSS)
—Command Injection
—Hardcoded Credentials
—Path Traversal
—Insecure Deserialization

</td> <td>

Code Quality

—Memory Leaks
—Race Conditions
—Null Pointer Dereference
—Off-by-One Errors
—Resource Leaks
—Infinite Loops

</td> </tr> </table>

Quick Start

python

from unsloth import FastLanguageModel

# Load model
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name="boraoxkan/codereview-ai",
    max_seq_length=2048,
    load_in_4bit=True,
)
FastLanguageModel.for_inference(model)

# Analyze code
prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
Analyze this Python code for defects.

### Input:
def get_user(username):
    query = "SELECT * FROM users WHERE username = '" + username + "'"
    cursor.execute(query)
    return cursor.fetchone()

### Response:
"""

inputs = tokenizer([prompt], return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.1)
result = tokenizer.decode(outputs[0])

Example Output

Input Code (SQL Injection vulnerability):

python

def get_user(username):
    query = "SELECT * FROM users WHERE username = '" + username + "'"
    cursor.execute(query)

Model Output:

json

{
  "code_quality_score": 20,
  "critical_issues": [
    "SQL Injection vulnerability due to direct string concatenation"
  ],
  "suggestions": [
    "Use parameterized queries to prevent SQL injection",
    "Handle database connections properly"
  ],
  "fixed_code": "def get_user(username):\n    query = \"SELECT * FROM users WHERE username = ?\"\n    cursor.execute(query, (username,))"
}

Score Guidelines

Score	Level	Description
0-30	Critical	Severe security vulnerabilities
31-50	Poor	Significant issues present
51-70	Fair	Some improvements needed
71-85	Good	Minor issues only
86-100	Excellent	Clean, secure code

Training

Parameter	Value
Dataset	~500 synthetic samples
Steps	120
Batch Size	1 (effective: 4)
Learning Rate	2e-4
Optimizer	AdamW 8-bit
Precision	BF16
Hardware	RTX 3070 (8GB)
Time	~40 minutes

LoRA Config

python

r = 16
lora_alpha = 16
lora_dropout = 0
target_modules = [
    "q_proj", "k_proj", "v_proj", "o_proj",
    "gate_proj", "up_proj", "down_proj"
]

Limitations

—Context limited to 2048 tokens
—Optimized for single-function analysis
—May produce false positives for complex patterns
—Training data is synthetically generated

Links

Resource	Link
GitHub Repository	boraoxkan/CodeReview
Base Model	Qwen2.5-Coder-7B
Unsloth	unslothai/unsloth

Citation

bibtex

@software{codereview_ai_2025,
  title = {CodeReview AI: Automated Code Analysis with Fine-tuned LLMs},
  author = {Bora Ozkan},
  year = {2025},
  url = {https://huggingface.co/boraoxkan/codereview-ai}
}

License

MIT License - See LICENSE for details.

<div align="center"> <b>Built with Unsloth & Qwen2.5-Coder</b><br> <sub>Making code reviews smarter, one bug at a time.</sub> </div>