mangsense/codebert_java

1---2library_name: transformers3pipeline_tag: text-classification4tags:5- text-classification6- pytorch7- jax8- code_x_glue_cc_defect_detection9- code10- roberta11- security12- vulnerability-detection13- codebert14- apache-2.015license: apache-2.016---17 18# CodeBERT fine-tuned for Java Vulnerability Detection19 20CodeBERT model fine-tuned for detecting security vulnerabilities in Java code.21 22## Model Description23 24This model is fine-tuned from [microsoft/codebert-base](https://huggingface.co/microsoft/codebert-base) for binary classification of secure/insecure Java code.25 26## Intended Uses27 28- Detect security vulnerabilities in Java source code29- Binary classification: Safe (LABEL_0) vs Vulnerable (LABEL_1)30 31## How to Use32```python33from transformers import AutoTokenizer, AutoModelForSequenceClassification34 35tokenizer = AutoTokenizer.from_pretrained("mangsense/codebert_java")36model = AutoModelForSequenceClassification.from_pretrained("mangsense/codebert_java")37 38# run code39```python40from transformers import AutoTokenizer, AutoModelForSequenceClassification41import torch42import numpy as np43tokenizer = AutoTokenizer.from_pretrained('mrm8488/codebert-base-finetuned-detect-insecure-code')44model = AutoModelForSequenceClassification.from_pretrained('mrm8488/codebert-base-finetuned-detect-insecure-code')45 46inputs = tokenizer("your code here", return_tensors="pt", truncation=True, padding='max_length')47labels = torch.tensor([1]).unsqueeze(0)  # Batch size 148outputs = model(**inputs, labels=labels)49loss = outputs.loss50logits = outputs.logits51 52print(np.argmax(logits.detach().numpy()))53```54 55## Training Data56 57Trained on CodeXGLUE Defect Detection dataset.58 59## Limitations60 61- Focused on Java code only62- May not detect all types of vulnerabilities