MhamedRayan commited on
Commit
6d2d9cd
·
verified ·
1 Parent(s): 3b6b7af

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +71 -0
README.md ADDED
@@ -0,0 +1,71 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: other
3
+ library_name: transformers
4
+ pipeline_tag: text-classification
5
+ base_model: microsoft/codebert-base
6
+ tags:
7
+ - codebert
8
+ - cybersecurity
9
+ - vulnerability-detection
10
+ - code
11
+ - security
12
+ - text-classification
13
+ ---
14
+
15
+ # Agent 2 — CodeBERT Vulnerability Classifier
16
+
17
+ CodeBERT-base fine-tuned to flag vulnerabilities in C/C++ source code.
18
+ Part of a three-agent DevSecOps auditor (metrics → transformer → gatekeeper).
19
+
20
+ ## Files
21
+
22
+ | File | Purpose |
23
+ |---|---|
24
+ | `config.json` | Architecture; `label2id` = `{"SAFE": 0, "VULNERABLE": 1}` |
25
+ | `model.safetensors` | Fine-tuned weights (~476 MB) |
26
+ | `tokenizer.json` | Fast tokenizer |
27
+ | `tokenizer_config.json` | Tokenizer settings |
28
+ | `meta.json` | Extra wrapper metadata (threshold, max length, metrics) |
29
+
30
+ `meta.json` is not used by `transformers`; the values are repeated here.
31
+
32
+ ## Specs
33
+
34
+ - Base model: `microsoft/codebert-base`
35
+ - Labels: `SAFE` (0), `VULNERABLE` (1)
36
+ - Max length: **384** tokens (longer input is truncated)
37
+ - Decision threshold: **0.45**
38
+ - Training data: CodeXGLUE defect detection + DiverseVul (C/C++)
39
+
40
+ | Metric | Value |
41
+ |---|---|
42
+ | F1 | 0.6881 |
43
+ | Precision | 0.5804 |
44
+ | Recall | 0.8450 |
45
+ | ROC-AUC | 0.7475 |
46
+
47
+ Recall is prioritized over precision, so false positives are expected. Scores
48
+ are softmax outputs, not calibrated probabilities. Non-C/C++ code is
49
+ out-of-distribution.
50
+
51
+ ## Usage
52
+
53
+ ```python
54
+ import torch
55
+ from transformers import AutoModelForSequenceClassification, AutoTokenizer
56
+
57
+ name = "your-username/agent2-codebert-vuln"
58
+ tok = AutoTokenizer.from_pretrained(name)
59
+ model = AutoModelForSequenceClassification.from_pretrained(name).eval()
60
+
61
+ inputs = tok(code, return_tensors="pt", truncation=True, max_length=384)
62
+ with torch.inference_mode():
63
+ logits = model(**inputs).logits
64
+
65
+ p_vuln = torch.softmax(logits, -1)[0, model.config.label2id["VULNERABLE"]].item()
66
+ vulnerable = p_vuln >= 0.45
67
+ ```
68
+
69
+ ## License
70
+
71
+ Not specified. Update the `license` field before publishing.