phamtungthuy/codebert-defect-detection

1---2language: en3license: mit4tags: [code, vulnerability-detection, defect-detection, codebert]5datasets: [devign]6base_model: microsoft/codebert-base7---8 9# CodeBERT for Defect Detection10 11Fine-tuned from [microsoft/codebert-base](https://huggingface.co/microsoft/codebert-base) for vulnerability detection.12 13## Usage14```python15from transformers import RobertaTokenizer, RobertaForSequenceClassification16import torch17 18tokenizer = RobertaTokenizer.from_pretrained("phamtungthuy/codebert-defect-detection")19model = RobertaForSequenceClassification.from_pretrained("phamtungthuy/codebert-defect-detection")20 21code = "int main() { char buf[10]; strcpy(buf, argv[1]); }"22inputs = tokenizer(code, return_tensors="pt", truncation=True, max_length=400)23prob = torch.sigmoid(model(**inputs).logits)24print("Vulnerable" if prob > 0.5 else "Secure")25```26