Text Classification
Transformers
Safetensors
code
roberta
code-classification
vulnerability-detection
automatic-vulnerability-detection
secure-coding
text-embeddings-inference
Instructions to use jacpacd/vuln-detector-codebert with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use jacpacd/vuln-detector-codebert with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-classification", model="jacpacd/vuln-detector-codebert")# Load model directly from transformers import AutoTokenizer, AutoModelForSequenceClassification tokenizer = AutoTokenizer.from_pretrained("jacpacd/vuln-detector-codebert") model = AutoModelForSequenceClassification.from_pretrained("jacpacd/vuln-detector-codebert", device_map="auto") - Notebooks
- Google Colab
- Kaggle
| license: mit | |
| language: | |
| - code | |
| library_name: transformers | |
| tags: | |
| - text-classification | |
| - code-classification | |
| - vulnerability-detection | |
| - automatic-vulnerability-detection | |
| - secure-coding | |
| # Vulnerability Detector for C Code (SARD) | |
| This model is a fine-tuned version of `microsoft/codebert-base` designed to detect vulnerabilities in C source code functions. | |
| ## Model Description | |
| This is a binary text-classification model that takes a C function as input and classifies it as either **Vulnerable** (`LABEL_1`) or **Safe** (`LABEL_0`). | |
| The model was specifically fine-tuned on the [NIST SARD (Software Assurance Reference Dataset)](https://samate.nist.gov/SARD/), focusing on common C vulnerabilities like Memory Leaks, Buffer Overflows, and other CWEs present in the Juliet Test Suite. Due to the clean and structured nature of the SARD dataset, the model achieved a very high accuracy on the validation set. | |
| ## Intended Uses & Limitations | |
| This model is intended as a proof-of-concept tool to assist developers in identifying potentially vulnerable code patterns during the development lifecycle. | |
| **Limitations:** | |
| * The model is highly specialized for the types of vulnerabilities found in the SARD dataset. Its performance on real-world, messy, or obfuscated code may be lower. | |
| * It should be used as an assistive tool, not as a replacement for comprehensive security audits or other static analysis tools. | |
| * The model classifies entire functions and may not pinpoint the exact line of code responsible for the vulnerability. | |
| ## How to Use | |
| The model can be easily used with the `transformers` library `pipeline`. | |
| ```python | |
| from transformers import pipeline | |
| # Load the classifier pipeline | |
| classifier = pipeline("text-classification", model="jacpacd/vuln-detector-codebert-c-sard") | |
| # Example of a vulnerable C function (Memory Leak) | |
| vulnerable_code = """ | |
| void CWE401_Memory_Leak__strdup_char_01_bad() | |
| { | |
| char * data; | |
| data = NULL; | |
| { | |
| char myString[] = "myString"; | |
| /* POTENTIAL FLAW: Allocate memory from the heap */ | |
| data = strdup(myString); | |
| printLine(data); | |
| } | |
| /* POTENTIAL FLAW: No deallocation of memory */ | |
| ; | |
| } | |
| """ | |
| # Example of a safe C function | |
| safe_code = """ | |
| void CWE401_Memory_Leak__strdup_char_01_goodB2G() | |
| { | |
| char * data; | |
| data = NULL; | |
| { | |
| char myString[] = "myString"; | |
| data = strdup(myString); | |
| printLine(data); | |
| } | |
| /* FIX: Deallocate memory */ | |
| free(data); | |
| } | |
| """ | |
| results_vuln = classifier(vulnerable_code) | |
| results_safe = classifier(safe_code) | |
| print(f"Vulnerable Code Prediction: {results_vuln[0]}") | |
| # Expected output: {'label': 'LABEL_1', 'score': 0.99...} | |
| print(f"Safe Code Prediction: {results_safe[0]}") | |
| # Expected output: {'label': 'LABEL_0', 'score': 0.99...} |