Bernadetta14 commited on
Commit
bcc0ded
·
verified ·
1 Parent(s): 722f23d

Delete code_vuln_app.py

Browse files
Files changed (1) hide show
  1. code_vuln_app.py +0 -140
code_vuln_app.py DELETED
@@ -1,140 +0,0 @@
1
- import gradio as gr
2
- from transformers import AutoTokenizer, AutoModelForCausalLM
3
- import torch
4
-
5
- # Load model
6
- MODEL_ID = "Bernadetta14/qwen2.5-coder-0.5b-vulnerability-detector"
7
-
8
- print("Loading model...")
9
- tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
10
- model = AutoModelForCausalLM.from_pretrained(
11
- MODEL_ID,
12
- torch_dtype=torch.float32,
13
- device_map="cpu"
14
- )
15
- model.eval()
16
- print("Model loaded!")
17
-
18
- SYSTEM_PROMPT = "You are a code security expert. Analyze the given code for security vulnerabilities and provide a structured security report."
19
-
20
- def analyze_code(code, lang):
21
- if not code.strip():
22
- return "⚠️ Please enter some code to analyze."
23
-
24
- prompt = f"""<|im_start|>system
25
- {SYSTEM_PROMPT}<|im_end|>
26
- <|im_start|>user
27
- Analyze this {lang} code for security vulnerabilities:
28
-
29
- ```{lang}
30
- {code[:800]}
31
- ```
32
-
33
- Provide a structured security report:<|im_end|>
34
- <|im_start|>assistant
35
- Vulnerability :"""
36
-
37
- inputs = tokenizer(prompt, return_tensors="pt")
38
- with torch.no_grad():
39
- outputs = model.generate(
40
- **inputs,
41
- max_new_tokens=300,
42
- do_sample=False,
43
- pad_token_id=tokenizer.eos_token_id
44
- )
45
-
46
- new_tokens = outputs[0][inputs["input_ids"].shape[1]:]
47
- result = tokenizer.decode(new_tokens, skip_special_tokens=True)
48
- return "Vulnerability :" + result
49
-
50
- # Contoh code vulnerable
51
- EXAMPLES = {
52
- "Python - SQL Injection": """\
53
- def login(username, password):
54
- query = "SELECT * FROM users WHERE username='" + username + "' AND password='" + password + "'"
55
- result = db.execute(query)
56
- return result
57
- """,
58
- "JavaScript - XSS": """\
59
- function displayMessage(userInput) {
60
- document.getElementById('output').innerHTML = userInput;
61
- }
62
- """,
63
- "PHP - Command Injection": """\
64
- <?php
65
- $filename = $_GET['file'];
66
- $output = shell_exec('cat ' . $filename);
67
- echo $output;
68
- ?>
69
- """
70
- }
71
-
72
- with gr.Blocks(title="Code Vulnerability Detector", theme=gr.themes.Soft()) as demo:
73
- gr.Markdown("""
74
- # 🔐 Code Vulnerability Detector
75
-
76
- Deteksi celah keamanan dalam kode secara otomatis menggunakan AI.
77
- Model akan menganalisis kode dan memberikan laporan terstruktur berisi
78
- jenis vulnerability, tingkat bahaya, penjelasan, dan saran perbaikan.
79
-
80
- **Model:** Qwen2.5-Coder-0.5B (fine-tuned on CyberNative Security Dataset)
81
- """)
82
-
83
- with gr.Row():
84
- with gr.Column():
85
- lang_input = gr.Dropdown(
86
- choices=["python", "javascript", "java", "c++", "c#", "php", "ruby", "swift", "go", "kotlin"],
87
- value="python",
88
- label="🌐 Programming Language"
89
- )
90
- code_input = gr.Code(
91
- label="📝 Code to Analyze",
92
- language="python",
93
- lines=15,
94
- value=EXAMPLES["Python - SQL Injection"]
95
- )
96
- with gr.Row():
97
- example_btn = gr.Dropdown(
98
- choices=list(EXAMPLES.keys()),
99
- label="📚 Load Example",
100
- value="Python - SQL Injection"
101
- )
102
- analyze_btn = gr.Button("🔍 Analyze Code", variant="primary")
103
-
104
- with gr.Column():
105
- output = gr.Textbox(
106
- label="📋 Security Report",
107
- lines=15,
108
- interactive=False
109
- )
110
-
111
- gr.Markdown("""
112
- ---
113
- ### 📊 Model Performance
114
- | Metric | Value |
115
- |--------|-------|
116
- | Accuracy | 62% |
117
- | Training Samples | 4,187 |
118
- | Supported Languages | 11 |
119
- | Model Size | 0.5B params |
120
-
121
- **Dataset:** CyberNative/Code_Vulnerability_Security_DPO
122
- **Method:** LoRA fine-tuning via MLX-LM (Apple Silicon)
123
- """)
124
-
125
- # Load example
126
- def load_example(example_name):
127
- return EXAMPLES.get(example_name, "")
128
-
129
- example_btn.change(fn=load_example, inputs=example_btn, outputs=code_input)
130
-
131
- # Update code language highlight when lang changes
132
- lang_input.change(fn=lambda l: gr.Code(language=l if l in ["python", "javascript", "java"] else "python"), inputs=lang_input, outputs=code_input)
133
-
134
- analyze_btn.click(
135
- fn=analyze_code,
136
- inputs=[code_input, lang_input],
137
- outputs=output
138
- )
139
-
140
- demo.launch()